Sapience Changelog For Developers

Table of Contents:

2026.10.19 Sharing > Database And Project Lists Show Items Collaborators Create

In a shared project or database, a table, view or folder that a collaborator created inside it could be missing from the database's lists of tables, views and automations, and from a project's one-level listing, for the owner and for people the whole project is shared with, even though they could open it. These lists now show exactly what you have access to, the same as the full project tree. People without access still do not see those items.

2026.10.19 Skills > Agents Say Which Skill Tools They Are Missing

When a skill you choose needs tools your agent does not have, the agent now tells you plainly which part of your request it cannot do and carries on with the rest, instead of working around the gap.

2026.10.19 Sharing > Public File Links Say When To Try Again

If a shared file cannot be loaded for a moment, its public link now says the file is temporarily unavailable and to try again, instead of saying the file was not found. A link to a file that has been removed still says the file was not found. For developers, the temporary case now answers with HTTP 503 rather than 404.

2026.10.18 Sharing > Shared Project Trees List Every Folder Collaborators Create

In a shared project, a folder that a collaborator created inside the project could be missing from the project tree for the project owner and for people the whole project is shared with, even though they could open it. The project tree now lists exactly what you have access to in the project, including the folders collaborators create and everything inside them. People without access to the project still do not see them.

2026.10.18 Databases > Automations Run The Database's Agent When Something Happens

A database's Actions menu now has Automations. Pick an event (a row is added, a row enters a view, a row is updated, a form is submitted or a row is deleted), choose the table or view it watches, and write what the database's agent should do. Each time the event happens, the agent carries out your instructions as you and records every run in one Automation conversation, which shows a lightning bolt in your conversation list and sits in its own Automations group. Only the person who created an automation can edit it; anyone else who can edit the database can switch it off or delete it, and people who can read the database can see it. Changes the agent makes during a run never start another automation, and a database can hold up to 25 automations.

2026.10.18 Databases > Email Everyone With Access When A Form Is Created

When someone creates a form on a database, everyone else who has access to that database now gets an email about it. It is on by default, and the Automations window has a switch to turn it off for that database. The same window also holds the existing switch that emails a form's owner each time the form is submitted.

2026.10.18 API > Database Automations Use The Existing Object And Tree Calls

Automations are a new Sapience object type, automation (uid prefix auto-), stored as a child of their database. Create, update and delete them with the existing /api/v2/sapience object calls using metadata_type "automation", and list them with POST /api/v2/srm/tree and types ["automation"]. The events an automation can listen for are published as the static value database-automation-events, read with GET /api/v2/sapience/by-external-key/database-automation-events. No new endpoints were added.

2026.10.18 Agents > EFQM Assessments Agent Assembles The Feedback Report

The EFQM Management Report agent is now the EFQM Assessments Agent, and it assembles the EFQM Feedback Report as well as the Management Report. Point it at an assessment's Feedback Report folder and it reads the assessors' feedback note, the team's scoring workbook and the closing-meeting presentation, then saves two files in that folder: the Feedback Report as JSON, and one HTML document that prints cleanly to PDF in A4 landscape. The feedback appears exactly as the assessors wrote it, the scores are recalculated from the scoring workbook and checked against the workbook's own totals, and the document has the same look as the Management Report. A folder whose note is still the unfilled template is refused rather than assembled, so a draft folder is never mistaken for the final report. Each assessment now has an Assessment record that links its Management Report and its Feedback Report, and a Management Report prepared outside Sapience can be recorded with a short description in place of a link. The EFQM report tools now run only in the EFQM Assessments Agent.

2026.10.18 API > Assemble An EFQM Feedback Report And Work With Assessments

POST /v2/utilities/tools/execute accepts three new tool names. "generate_efqm_feedback_report" takes a folder_uid and a required assessment_uid, writes the Feedback Report JSON and HTML into that folder, and links the report to the assessment; given a project uid instead of a folder, it assembles nothing and returns the candidate Feedback Report folders. "list_efqm_assessments" returns your assessments with their uid, name and creation date, and "create_efqm_assessment" creates one. Generating a Management Report now also records the assessment it belongs to, creating one when the folder has none.

2026.10.18 Menu > The Collapsed Sidebar's Expand Button Pins It Open

Clicking the expand arrow on the collapsed sidebar now always brings the full sidebar back and keeps it open, even if you pause over the arrow first.

2026.10.18 Notes > Block Menu For Expansion Items, Tables And Images

Hover any expansion item, table, image or embed in a note to reveal a handle on its left, or right-click it, to delete, copy or cut the whole block, or insert a blank line after it. Backspace after a block now selects it, and a second Backspace deletes it.

2026.10.18 Notes > Actions Menu In The Note Editor

The note editor now has the same Actions menu as the note viewer, next to Save, and shows the note's key details without leaving the editor.

2026.10.18 Skills > Printouts Always Come With A Shareable Link

Asking an agent for a printout now always returns a web link you can open and share, not just the saved file.

2026.10.18 Agents > Faster File Uploads To New Agents

Adding several batches of files to a new agent at the same time now answers promptly for every batch.

2026.10.18 Menu > Search Results In Regular Weight

Result names in the search window used to all show in bold, so no result stood out. They now show in regular weight, which makes the list easier to scan. The result count and the filter and type chips are unchanged.

2026.10.18 Notes > Comments Load Faster On Notes, Files And Pages

Comments now load faster across Sapience. The comments list that fills in when you open the app, the unread comment counts, and the comments shown on a note, file or page all arrive sooner, especially when you can see comments on many items.

2026.10.18 Agents > Home Shows The Full Agent Welcome When You Pick An Agent

Picking an agent on the Home screen now shows the same agent banner you see in chat: the agent's avatar, its full description and an About this agent link. Below it, Home shows the agent's complete welcome message, including the agent's email address when it has one and its disclaimer, and the starter questions use the same chips as chat. Clear Agent/Skill Combo now sits right beside the agent picker.

2026.10.18 Chat > Keep The Sidebar Open

After collapsing the sidebar, clicking the expand arrow now keeps it open, and it stays open after a reload on that device. Hovering the collapsed sidebar still previews it.

2026.10.18 Databases > Simpler Table Tabs

Table tabs are now compact, so a database's tables fit in one row without scrolling, and you reorder them by dragging a tab itself. The Views menu's table headings now use the same tinted headings as the Actions menu.

2026.10.18 Chat > Opening A Chat Recovers From A Brief Network Drop

If the connection dropped for a moment while a conversation was opening, the chat could show a message saying the conversation details could not be refreshed. Opening a conversation now retries automatically and loads normally after a brief drop. If the connection stays down, the message still appears so you know to try again.

2026.10.17 Projects > Project Admins Can Remove A Member's Item From A Project

A project's owner or admin can now take an item that another member added out of the project with Remove from Project. The item moves back to its owner's own files instead of being deleted, so nothing is lost, and only an item's owner can delete it.

2026.10.17 Agents > Weather Questions Use Live Web Research

Weather questions are now answered with live web research.

2026.10.17 Notes > Embed A Live Database Table Or View In A Note

Type /database in a note to open Insert a database, which lists your databases the same way the Everything view does. Expand a database to see its tables and views, then pick one to drop it into the note as a block. The block shows that table's or view's live rows read-only with a View only badge, keeps working after you save and reopen the note, and appears in the note viewer with an Open in Databases link that takes you straight to that database, table and view. If the database is later deleted or you lose access, the block says it isn't available instead of breaking the note.

2026.10.17 Databases > Linked Rows Open The Real Project, Note, File Or Folder

A reference chip on a linked database row now names the kind of object it points at, and clicking a chip or a linked row opens that project, note, file or folder directly instead of just the row. Linking an existing item into a table now offers your projects and folders alongside files and notes, so a table can point at anything you work with.

2026.10.17 Databases > Duplicating Without Records Leaves Notes Behind

Duplicating a database without its records used to copy the notes stored inside it along. Now the empty copy arrives with the tables, views and form only, while copies made with the records still bring their own editable copies of every note and file.

2026.10.17 Email > Agent Emails Arrive From The Agent

Emails sent by an agent now arrive from that agent's own address, so replying writes back to the same agent instead of a generic no-reply mailbox. Each message also carries a short footer naming the conversation it belongs to. If an agent has no dedicated address, its mail keeps using the existing sender exactly as before.

2026.10.17 Chat > Favorites Stay At The Top Of The Conversation List

The sidebar conversation list used to show the Projects group above Favorites. Favorites now always come first, followed by Projects whenever a project holds a visible conversation, then the remaining groups in their usual order.

2026.10.17 Projects > Chats About A Project's Files Appear In Its Conversations

A project's Conversations tab now also lists chats about a file, folder or note that lives in that project, alongside the project's own chats. A Type column shows what each chat is about, and a Linked Source column opens the file, folder or note behind it. Moving a note to another project moves its chat with it.

2026.10.17 Databases > Views Menu Groups Views By Table

The Views button read All rows even when no view was open; it now reads Select View, while the All rows item still clears the view. Views are also grouped under one header per table in table order, so each table's views sit together. Databases with views on only one table look as they did.

2026.10.17 Databases > Clearer Actions Menu With Help For Each Database

The database Actions menu, on the table header and on right-click, now names each action plainly (New Table, Rename Database, Duplicate Database and the like), groups items under tinted headings, and explains every item on hover. Databases that document themselves also offer a Help group: a How to Use guide that opens in the app, plus links to the database's help and website pages. Empty or missing help details simply show no entry.

2026.10.17 Databases > Changing A Column's Type Reports What Was Kept

Changing a column's type from the Edit-column dialog now confirms what happened: a message reports how many cell values were kept and how many were cleared. The message appears once, where the change was made.

2026.10.17 Notes > Pressing Enter In An Expansion Title Stays In Its Item

With two expansion items open back to back, pressing Enter at the end of the first title could drop the new line and the caret into the second item's body. Clicking into a title now syncs the editor position first, so the new line lands in the item you pressed Enter in.

2026.10.17 Notes > Turning Lines Into Expansion Items Keeps Their Text

Converting a line or heading into an expansion item keeps the line's text as the title (headings keep their level), and the toggle arrow is larger and centred on the title's first line in both editor and viewer. A list line followed by a code block or heading also converts cleanly instead of breaking the note's structure.

2026.10.17 Desktop > Double-Click The Tray Icon To Open Sapience

Double-clicking the Sapience tray icon now opens the app on every platform. The tray menu item previously called Open Dashboard is now called Open Sapience.

2026.10.17 Menu > Press Alt+R To Report An Issue

Pressing Alt+R (Option+R on macOS) from anywhere in the app now opens the Report an Issue modal, including from inside a note. It behaves exactly like opening it from the menu, stays open if pressed again, and is listed in the Keyboard Shortcuts panel.

2026.10.17 Files > Larger Spreadsheet Parts Now Open

Very large spreadsheets whose parts the document reader used to reject are now accepted, with parts up to 200 MB supported. When a file still cannot be read, Sapience remembers that and says so straight away, suggesting you split the sheet or export it as CSV, instead of retrying the read each time you open it.

2026.10.17 Outbound > Set A Sender's Daily Limit From The Senders Page

The Senders page in the Outbound Ops Console now has a Limits button for each sender. It opens a Sender limits panel where you set Total sends/day, the most invites, LinkedIn messages and emails that sender may send in one day across every campaign, then Save settings. Leave it blank and the panel shows the default limit the sender gets instead. On a campaign's Sending & schedule tab, the Daily cap box sets that same sender-wide total, and now says so: a blank box used to read "no cap" and now shows the default limit that applies, hovering over it explains what it limits and when the count resets, and an Edit sender limits link opens the Senders page.

2026.10.17 Outbound > Warm-Up Never Raises A Sender Above Its Daily Limit

A sender in warm-up with no daily limit of its own could be allowed more sends a day than the default limit once its ramp passed that number. Warm-up now only ever lowers a sender's daily limit: the sender's own limit, or the default when none is set, is always the ceiling. The Outbound Ops Console's reason for why a sender is waiting now uses the same limits as the sending engine, so what it shows matches what will happen.

2026.10.16 Notes > Copy A Link To A Note

Notes now have a Copy link action in the note list's right-click menu, the note viewer's More menu and the editor's File menu, on desktop and mobile. It copies a link that opens that exact note in Sapience. In the desktop app the link uses the public web address, so anyone you send it to can open it in their browser.

2026.10.16 Notes > Room To Keep Typing Below An Image

When an image was the last thing in a note, there was no line below it to click into. The note editor now always keeps an empty line after an image that ends a note or sits directly before another block such as a table, so you can click below it and keep typing. Opening a note does not count as an edit, and undoing an image you just added removes its empty line too.

2026.10.16 Settings > Start A Floating Chat On The Dashboard

A new switch in Menu > Settings & Customization > Advanced, Dashboard: Start Floating Chat by Default, opens a floating chat with the Sapience Agent every time you land on the Dashboard, ready to type in. Leave the Dashboard and come back, and the same conversation is waiting. The switch is off until you turn it on, follows your account to every device, and does not apply on phones.

2026.10.16 Chat > Replies Arrive When The Live Connection Is Down

Extended the live-connection handling for agent replies. If the live connection is down or goes quiet after you send a message, Sapience now checks for the agent's reply over a standard request and shows it as soon as it is ready, rather than waiting for the connection to come back, and a reply is never shown twice.

2026.10.16 Agents > Agents Carry On When A Tool Is Not Available

Hardened agent tool calling for the case where an agent asks for a tool that is not available in the current step. The agent now gets a clear note that the tool is not available, with the correct name when it simply used the wrong form of one, and carries on with the tools it has to finish your request.

2026.10.16 Databases > Resize Table Columns By Dragging

In a database's table view, drag the border between two column headers to make a column wider or narrower. The width is saved on the open table view and restored the next time anyone opens it. Widths are saved for people who can edit the view; for everyone else a drag lasts for the visit. Autosize in the column menu works as before.

2026.10.16 Projects > Project Admins Can See And Change Member Access Levels From The Team Tab

The Team tab of a project now shows project admins each member's access level and lets them switch an ordinary member between Read and Write directly from the list, without asking the owner. The owner, other admins and your own row are shown as fixed, and only the project owner can grant the admin role.

2026.10.15 Projects > Project Responses Include Each Member's Access Level

Reading a project now returns member_access_levels, a map of each member's username to their current access level: owner, admin, write or read. It is shown to the same people who can already see the project's member list, and it updates whenever a member's level is changed. The single project read always returns fresh levels; the shorter project rows in tree and list views carry them once that project's membership has next been updated.

2026.10.15 Databases > Edit Column Settings For Select And Multi-select Columns

Right-clicking a Select or Multi-select column header now offers Edit column settings, where you can add, delete and rename options and pick each option's colour, both when you create the column and at any time later. Colours stay exactly as you set them, and light colours get dark text so labels stay readable. Renaming an option updates every row that held the old name and any saved view that filters on it, while deleting one option and adding another leaves existing rows alone. Multi-select cells now open an option picker with one click, and a cell with many values shows the first two as coloured chips plus a +N chip that lists the rest.

2026.10.15 Databases > Delete A Saved View

A Delete view button now sits beside Edit view when a saved view is open. It asks for confirmation, removes only that view (your rows stay), and returns the table to All rows. It is shown to the database's owner, to people with admin access on its share, and to superusers; people with read-only access do not see it.

2026.10.15 Databases > Redesigned Create And Edit View Dialog

The Create view and Edit view dialog has a cleaner layout: Name and Layout sit side by side, and the sections follow in the order Filter, Sorts, Group by and Visible columns, separated by light dividers. Match all or Match any is a compact toggle on the Filter heading, conditions and sorts use compact rows with Add condition and Add sort links, visible columns are clickable chips, and a greyed-out Save shows why it is disabled. The is one of operator takes a single comma-separated text field, and values that themselves contain a comma are shown in quotes so they keep their meaning when edited.

2026.10.15 Databases > Redesigned New Form Build Tab

The Build tab of the New form dialog has a lighter design. Add field is a compact button that opens a dropdown of the table's available columns, with column types that can't be used in a public form shown greyed out and locked in that dropdown instead of on the page. Field cards, option rows, the Submit button text and the resizable Thank-you message all share the same input style, and Settings are a flat list with each label on the left and its switch on the right.

2026.10.15 Databases > Choose Reference Rows Shows Names With A Row Info Panel

The Choose reference rows dialog now lists each row as a single clean line with an always-present checkbox and the row's name. A small info button on each row opens a panel with every column of that row, formatted the way the table shows it, including dates in your time zone and formatted currency, even when the linked table lives in another database. The search field has a Clear button inside it, and the footer shows how many rows are selected.

2026.10.15 Databases > Table Area Scrolls Inside The Window At Any Zoom

At high browser zoom, the bottom of a shorter table, including its left-right scrollbar and + Add row, could sit below the window. On desktop, the Table and Pivot area now scrolls inside the window, so the scrollbar, the last rows and + Add row are always reachable, and a short table shows no extra scrollbar. Board, Gallery, List and Calendar layouts and phones are unchanged.

2026.10.15 Projects > A Note When Only Active Items Are Shown

The Projects, Databases and Notes pages now show a short note at the top of the list while the Active filter is applied, explaining that templates, archived items and other statuses are hidden and that the view filter on the right shows everything. Choosing any other filter hides the note, and it is available in every app language.

2026.10.15 Projects > Save To Project Offers The Item's Own Project

When you chat about a note, file, folder or database that lives inside a project, Save to Project now asks whether to save the chat to that project or choose a different one. Picking the current project saves straight there with no picker; any other chat opens the project picker as before.

2026.10.15 Sharing > Team Tab Explains Members An Admin Cannot Remove

When a project admin ticks the owner or another admin along with ordinary members and clicks Revoke Access, the ordinary members are still removed and a message now says how many were skipped, because a project admin can only remove ordinary members.

2026.10.15 Chat > Replies Fetched On Wake Show Once With Their Plan And Files

Extended the sleep-and-wake handling for agent replies to the case where the app has already fetched the missed answer by the time it wakes. The chat now treats that answer as delivered: it shows a single copy, with no extra empty reply bubble and no failure note, and still updates the plan and any files the agent produced during that turn.

2026.10.14 Agents > Sapience Chief People Officer Agent

Your AI head of people. With no HR team, it acts as one: people strategy tied to your business plan, an annual people calendar, and lawful hiring, pay, and exit policies. If you already lead HR, it is your chief of staff: frameworks, trade-offs, and drafts off your desk. Five specialists cover hiring, onboarding, performance, pay, and employee relations; it returns one answer in one voice. It covers Australia, the United States, and the United Kingdom in depth, and elsewhere gives the method plus a local-law flag. Find it on the People & Talent shelf of the AI Store; the Chief of Staff agent now sends people questions here.

2026.10.14 Agents > Sapience Talent Acquisition & Candidate Assessment Agent

Hire on evidence, not gut feel. This agent designs role scorecards, inclusive job ads, and structured interviews with anchored rating scales, then scores every candidate criterion by criterion against quoted evidence from your notes. It never ranks candidates and never tells you who to reject: every hiring decision stays with a person. It keeps your hiring lawful across Australia, the United States, and the United Kingdom, including the guardrails around AI screening tools, and elsewhere gives you the method and flags what to check locally. Find it on the People & Talent shelf of the AI Store.

2026.10.14 Agents > Sapience Onboarding & Early Success Agent

Turn new hires into productive, compliant, connected employees, fast. This agent builds the preboarding sequence, the day-one compliance checklist for your country, 30-60-90 day plans, manager onboarding for first-time people leaders, and probation done lawfully, and it helps you catch early-tenure wobbles before they cost you the hire. It covers Australia, the United States, and the United Kingdom in depth, and elsewhere gives you the method and flags what to check locally. Find it on the People & Talent shelf of the AI Store.

2026.10.14 Agents > Sapience Performance & Development Agent

Reviews people actually trust, goals people actually hit, and improvement plans that genuinely improve. This agent designs review cycles, goals and OKRs, turns messy manager notes into balanced evidence-based reviews with a shown bias check, runs calibration, builds career ladders and promotion packets, and drafts improvement plans with measurable objectives and review dates. The moment underperformance becomes a disciplinary matter it hands over to the Employee Relations specialist. It covers Australia, the United States, and the United Kingdom in depth. Find it on the People & Talent shelf of the AI Store.

2026.10.14 Agents > Sapience Workforce Planning & Rewards Agent

Tie headcount to money. This agent builds headcount plans bound to your budget and revenue plan at fully loaded cost, designs org structures with sane spans and layers, and sets levelling, pay bands, and benchmarks. It covers employee equity, pay equity audits, minimum-pay coverage, and contractor-versus-employee classification across Australia, the United States, and the United Kingdom, and elsewhere gives you the method and flags what to check locally. For final tax positions it sends you to a tax professional. Find it on the People & Talent shelf of the AI Store.

2026.10.14 Agents > Sapience Employee Relations & Investigations Agent

Run complaints, investigations, discipline, dismissal, and redundancy like a senior employee-relations specialist. This agent walks you through the first hours of a complaint, plans and pressure-tests workplace investigations, sequences dismissals for procedural fairness, and runs genuine redundancies with consultation and redeployment. Where claims, whistleblowers, criminal allegations, or settlements appear, it stops and sends you to a lawyer. It covers Australia, the United States, and the United Kingdom in depth, and the General Counsel agent now sends HR-process questions here. Find it on the People & Talent shelf of the AI Store.

2026.10.14 Sharing > Project Admins Can Remove Ordinary Members

Project admins can now remove ordinary members from a project, using the same Revoke Access action on the Team tab that project owners use, and the removed member leaves the list straight away. Through the API, a project admin can also change an ordinary member's access between read and write. Admins cannot remove or change the project owner, another project admin or themselves, and only the owner can grant the admin role.

2026.10.14 Chat > Automatic Reply Mode Only Moves Up From Minimal-Fast

When the reply mode is chosen for each message, a conversation now only moves up from Minimal-Fast, never back down to it. The first message of a chat can still be answered in Minimal-Fast; once any message in it has gone in Balanced or Think Longer, later messages go in Balanced at least, and can still move freely between Balanced and Think Longer. This also holds for messages typed while a reply is still arriving, for retries and edited resends, and after the page is reloaded in the same browser. A mode you choose yourself still always wins.

2026.10.14 Databases > File Fields Show Each File On Its Own Full-Width Row

In the New Row and Edit Row dialogs, a Files field now lists each attached file on its own full-width row, with the name and type on the left and the remove button at the right end. Images show a small rounded thumbnail inside the field, and people who can only view the row no longer see an empty divider under the list.

2026.10.14 Projects > File, Folder And Note Chats Appear In Their Project

A project's Conversations tab now also lists chats about a file, a folder or a note that lives in that project, directly or inside one of its folders, alongside the project's own chats. A Type column shows whether each is a project, file, folder or notes chat, and a Linked Source column opens the file, folder or note it is about. When a note moves to another project its chat moves with it, and clicking a row opens the chat with its own agent.

2026.10.14 Databases > Risk And Compliance Register Template

A new Risk & Compliance Register database template records your risks scored by likelihood and impact with their mitigation plans, the controls that address them, and a log of what actually went wrong, with example entries in every table. It comes with 14 saved views, including a risk heat map that counts risks by likelihood and impact, a board of active risks by status, high and critical open risks, overdue reviews, a mitigation calendar, controls that are not working, and the open entries in the log with their cost by severity, plus a form for reporting a new risk or problem. Find it under Templates on the Databases page, choose Duplicate to get your own editable copy, and untick Copy records too to start empty.

2026.10.14 Databases > Hiring Template

A new hiring database template tracks the roles you are filling, the candidates who applied and the interviews your panel ran, with example entries in every table. It comes with 13 saved views, including open roles by start date, a candidate board by stage, top-rated candidates still in the running, referrals, a sourcing pivot, new arrivals and an October interview calendar, plus an Apply for this role form. Find it under Templates on the Databases page, choose Duplicate to get your own editable copy, and untick Copy records too to start empty.

2026.10.14 Databases > Customer Support Desk Template

A new Customer Support Desk database template tracks your customers and their plans, the requests they send with priority, status and time spent, and the responses your team sent back, with example entries in every table. It comes with 15 saved views, including a request board by status, urgent and high requests still open, unassigned requests, VIP customers, an October calendar, satisfaction and time-spent pivots and a phone log, plus a Get help form. Find it under Templates on the Databases page, choose Duplicate to get your own editable copy, and untick Copy records too to start empty.

2026.10.14 Databases > Contracts And Suppliers Template

A new Contracts & Suppliers database template tracks your suppliers, their contracts with values in US dollars and renewal dates, and periodic supplier reviews scoring quality, delivery and service, with example entries in every table. It comes with 15 saved views, including preferred suppliers, a board of contracts by status, high-value active contracts, auto-renewing contracts, upcoming renewals, a renewal calendar, spend and contract-value pivots and reviews that need action, plus a New supplier request form. Find it under Templates on the Databases page, choose Duplicate to get your own editable copy, and untick Copy records too to start empty.

2026.10.14 Databases > Content And Marketing Calendar Template

A new Content & Marketing Calendar database template plans your campaigns with goals and budgets in US dollars, follows every piece of content from idea to published, and records the results each piece brought in, with example entries in every table. It comes with 16 saved views, including an editorial board, content and campaign calendars, an ideas backlog, top performers, live campaigns, and leads-by-month and budget-by-goal pivots, plus a Suggest a content idea form. Find it under Templates on the Databases page, choose Duplicate to get your own editable copy, and untick Copy records too to start empty.

2026.10.14 Databases > Employee Directory Template

A new Employee Directory database template lists your departments, your offices and your people, each person with a photo, role, manager, start date, work arrangement and weekly hours, with example entries in every table. It comes with 14 saved views, including a team gallery of photos, a start-date calendar, a phone list, new starters, contractors, the remote team, people on leave and an hours pivot, plus a New starter details form. Find it under Templates on the Databases page, choose Duplicate to get your own editable copy, and untick Copy records too to start empty.

2026.10.14 API > Webhooks For Every Database Event

Outbound webhooks can now subscribe to four more database events, so all seven are available. database_row_removed fires when a row is removed from a table and carries the target_uid and target_type of the object it pointed at, which is kept. database_row_changed fires for every saved edit to a row, including a move or a rename, and names the changed columns without their values. database_schema_changed fires when a table's columns are added, edited, deleted or reordered. database_form_submitted fires when someone submits one of your public forms and carries the form title and each answer as a label and value, never the form owner's email address. Each can be narrowed to one table with a database_trigger that names only table_uid, and GET /api/v2/webhooks/eligible-events now lists fifteen events.

2026.10.14 API > Realtime Updates For Every Database Event

Every database event now reaches open apps over the realtime connection: a row's cells being updated, a row entering a saved view and a public form being submitted join rows being added, removed and changed and a table's columns changing. Each goes to the database's owner, the person who made the change, everyone the database is shared with and, for an organisation database, the organisation, and carries IDs only, so the app fetches the row through the usual access checks. A row entering a saved view is now announced for every saved view of the table, not only views a webhook watches, except views filtered on a row's name, creation time, position or type. The three new event types can also be read back through POST /api/v2/srm/tree with types events, and are in the TypeScript SDK's realtime event types.

2026.10.13 Chat > Each Open Chat Keeps Its Own Message Draft

With several chats open side by side, text typed in one chat's message box also appeared in the others. Each chat now keeps its own draft: typing in one leaves the others untouched, sending sends only that chat's text and clears only its box, and a minimised chat still holds what you typed when you open it again. An unsent draft is still not kept across a page reload.

2026.10.13 Files > A Repeat Upload Lands Where You Send It

Uploading a file you had already uploaded a few minutes earlier, to a different note, folder or project, could hand back the earlier copy instead of adding the file where you asked, so an attachment to a note might not appear in it. A repeat upload now lands where you send it. Only a repeat of the same upload to the same place is treated as a retry and answered with the copy you already have, so a retry over an unsteady connection still does not leave a duplicate copy.

2026.10.13 Files > Videos Whose Audio Can't Be Read Answer Straight Away

Hardened opening a chat on a video whose sound track can't be read. After the first attempt, every later try now gives the same answer straight away instead of working through the whole video again. Re-processing the file, or uploading a new version of it, tries again from the start.

2026.10.13 Agents > Accented File Names Open Reliably In An Agent's Workspace

Hardened how the files you give an agent are placed in its workspace (run_in_sandbox). Files whose names contain accented letters, including names saved on a Mac, are now placed under the same spelling the agent is told to use, so it opens them by name on the first try. The file keeps its original name everywhere else in Sapience.

2026.10.13 Files > Recordings Still Being Transcribed Say So When You Open Their Chat

Opening the chat on an audio or video file that is still being transcribed now shows one short message saying the file is still being processed and to try again in a moment, instead of an error saying the conversation could not be started. Once transcription has finished, the file's chat opens as usual.

2026.10.13 Email > Agent Emails With Large Attachments

Hardened agent emails that carry large attachments. An email with large attachments is now sent directly through a route that accepts larger messages, and one whose attachments are too large to send at all is turned down straight away with a message telling the agent to share a link to the file instead of attaching it, so the agent can send that follow-up email right away.

2026.10.12 Agents > Workspace Runs Stop At Their Time Limit Instead Of Starting Over

When a task in an agent's workspace (run_in_sandbox) reached its 20-minute limit, it could be started again from scratch and run for another 20 minutes before reaching the same limit. A run that reaches its time limit now ends there, so the agent gets the advice to split the work into smaller runs after 20 minutes rather than 40. A run that fails quickly for another reason still gets one automatic retry.

2026.10.12 Sharing > Project Collaborators Can Delete What They Can Edit

A collaborator with write access to a shared project could open and edit a task inside it but not delete it. Deleting now follows the same project-level access as editing, so write and admin collaborators can delete tasks, and empty goals, folders and pages, anywhere inside the project, and a project owner can delete a collaborator's task in their own project. Collaborators with read-only access still cannot delete.

2026.10.12 Menu > Browse All Your Templates In One Place

The Manage section of the menu now has a Templates item. It opens one list of every project, note and database that has been marked as a template, grouped by who it belongs to: you, your organization, or everyone on Sapience. All three groups open expanded and show how many templates they hold. Search matches a template's name, description, type or group, Refresh reloads the list, and clicking a row opens that template and closes the window. Descriptions keep their formatting, such as bold text, on a single line, and a long one shows in full when you hover over it.

2026.10.12 Agents > Agents Outside A Project Are Only Offered Tools They Can Use

An agent running outside a project, including a scheduled job, is no longer offered project-only tools such as searching a project's files and notes, so a run no longer spends steps on tools that need a project and its reply no longer explains that they were unavailable. Agents working inside a project keep every project tool.

2026.10.12 Projects > A Refused Task Delete Says Why

When a task in a project cannot be deleted, the message now gives the reason in one plain sentence instead of technical error text, and stays on screen until you dismiss it. The task stays in its place in the list. Deleting several tasks at once shows the reason for any that could not be deleted.

2026.10.11 Models > Automatic Reasoning Effort Tops Out At High

When Sapience chooses the reasoning effort for a turn automatically, it now picks low, medium or high on every model, and never moves a turn up to the slower extra-high tier on its own. Extra-high still applies when you choose it yourself, when a skill asks for it, and while an agent drafts a plan.

2026.10.11 Mobile > Agent Taps Keep Working While A Chat Is Still Opening

Hardened agent taps on phones. A tap made while the app is still starting up now waits for it to finish before opening the agent's chat, and if a chat is still opening after 15 seconds, the next tap takes over and opens its own chat instead of being ignored. A slow chat still opens in the agent you tapped, without creating a second chat. Desktop is unchanged.

2026.10.11 Databases > The Sapience Agent Builds A Database From A Conversation

Ask the Sapience Agent for a new database, or pick the Create a Database skill, and describe what you want to keep track of in your own words. The agent asks two or three short rounds of plain questions, each with an example answer, then shows a plan in everyday language: the lists it will make, the details on each, how the lists link together, the saved views, any forms, and the example entries. Nothing is built until you say yes. It then creates the whole database for you, with the right kind of detail for each piece of information (money in your currency, dates, yes or no, choices, ratings, lengths of time, links and attachments), links between related lists, details shown from linked entries, saved views with filters, sorting and grouping, forms where other people fill things in, and a few example entries in every list. When it finishes it tells you what it built, what each view and form is for, and where to open it.

2026.10.11 Agents > Database Agents Build Richer Views, Forms, Tags And Linked Entries

A database's own agent can now build everything a full database uses. It can tag the database, its lists, saved views and entries; create saved views that group rows, show only chosen columns, carry a short description and open as the list's default; add an existing note or file to a list as an entry with its own details, such as a summary or a link to another entry; and create forms with field labels, help text, required fields, an introduction, a button label and a thank-you message. A form the agent creates gets a public link only when you ask for one, optionally with a password. Lookup columns show a plain column of the linked list or the linked entry's name, and the agent is told clearly when a lookup it tries would point at another link or lookup.

2026.10.11 Agents > Agent-Created Checkbox Filters Match Correctly

A saved view an agent created to filter a checkbox column, for example to show only unpaid entries, could come back empty. Checkbox filters created by an agent now match ticked and unticked rows correctly.

2026.10.11 API > Tag Database Entries Through The Tag Endpoint

POST /v2/srm/tag now accepts database entry ids alongside notes, files, projects and tables, so entries can be tagged and untagged through the same call with no change to its request or response shape. Renaming or removing a tag everywhere also reaches the database entries that carry it.

2026.10.11 Databases > Board, Gallery, List, Calendar And Pivot Views

Every database table and saved view can now be shown six ways from the new Layout dropdown in the toolbar: Table, Board, Gallery, List, Calendar and Pivot. Board puts each row on a card in a lane for each value of a select, checkbox, person or rating column, in that option's colour, and dragging a card to another lane updates the row. Calendar places rows on their date in month, week and day views, and dragging an item to another day moves its date while keeping any time of day. Gallery shows a card per row with a cover image from a file column, List shows compact rows, and clicking either opens the row. A view you can edit keeps its layout and its settings for everyone, and the Demo Database now has a view of every type to try.

2026.10.11 Databases > Pivot Views Saved With Your View

Pivot is now a view type of its own. A Pivot view opens as a ready cross-tab, with one column's values down the side, another's across the top, and summed, averaged, counted, minimum or maximum values in the cells, built from the rows, columns and values saved with the view. If you can edit the view, rearrange the pivot in the side panel and choose Save pivot to keep it for everyone who opens it. The database area also always leaves room for the full pivot side panel, even on a table with only a few rows.

2026.10.11 Databases > Edit A Cell With One Click

Text and number cells now open for editing on a single click instead of two. In a single-line cell Enter saves and moves down to the next row, a long-text cell opens with its text selected and saves with Ctrl or Cmd plus Enter, and a small chevron expands a long cell so clicking the cell itself no longer resizes the row.

2026.10.11 Projects > A Refused Role Change Shows A Plain Message And Keeps The Previous Role

When a change to someone's role in a project's Shares tab is refused, for example because the project owner's role cannot be changed, the message is now one plain sentence instead of technical error text, and the role picker goes back to the role the person already had. Cancelling the confirmation also leaves the role picker as it was.

2026.10.11 Files > Uploads Where Every File Fails Now Say Why

When none of the files in an upload can be added, the message now gives the reason, for example that a file with the same name is already in that folder, instead of only saying the upload failed. When the files fail for different reasons, it shows the first reason and how many other files failed.

2026.10.10 Sharing > Not Shared And Not Published Use The Shared Neutral Badge

The Not shared and Not published labels on a note, and Not published on a project, now use the same grey neutral badge as the app's other status labels instead of an outlined chip. Clicking them still opens sharing and publishing as before.

2026.10.10 Agents > AI Agents Store Install Buttons Match The Rest Of The App

In the AI Agents Store, every Install button is now the app's small blue main button in place of a grey chip, with the same words and behaviour, and the selected category in the left-hand list is shown as a pale-blue row with blue text instead of a solid blue block.

2026.10.10 Meetings > Long Meetings Keep Their Recording Links

Hardened meeting capture for long meetings. Saving a long meeting's recordings now has up to thirty minutes to finish, so the meeting note keeps its links to the audio and video recordings alongside its transcript and summary.

2026.10.10 Notes > Notes Lists Use The Shared Folder Colour And The Project Icon

Folders and projects in the Notes list, grid and card views now use the same blue as Files and Projects in place of yellow and bright blue. Projects show the project icon from the menu instead of a folder, so folders and projects stay easy to tell apart, including in the card view's legend.

2026.10.09 Menu > One Shared Look For Badges, Buttons And Focus

The app's dark look now uses one shared set of colors. Status badges come in five fixed types (success, info, warning, danger and neutral), primary and secondary buttons share one blue style, the Projects header banner is a plain header like the other screens, and keyboard focus shows one 2px ring that never appears on a mouse click. Layout, positions and navigation are unchanged.

2026.10.09 Files > Image Descriptions Recover After An Update

Hardened image descriptions in documents: if Sapience is updated while it is describing a document's images, those images are described again on the next attempt.

2026.10.09 Agents > Web Research Agent Carries On When A Search Step Returns Nothing

Hardened the Web Research Agent so a search step that returns no results no longer ends the turn. The turn carries on and answers, without the sources from that empty step.

2026.10.09 Sharing > Zero-Day Link Expiry Gives A Clear Message

When an agent publishes a file as a public web link and asks for an expiry of zero days or less, it now receives a plain message that the expiry must be positive.

2026.10.09 Accounts > Sign-Up Passphrases With Accented Characters Are Checked Normally

Hardened sign-up passphrase checking for passphrases that contain accented letters, curly quotes or a non-breaking space, which are now checked normally and give the usual not-valid message when they do not match.

2026.10.09 Agents > Agents Describe All Their Abilities When Asked

An agent that looks up its tools as it needs them now answers what can you do from its whole list of abilities. Before, it could describe only the tools it had loaded at that moment and sometimes said it could not do something it could.

2026.10.09 Chat > Attach Up To Five Images To One Message

You can now paste or attach up to five images in one message in a conversation. The composer shows each image in the order you added it, with a count and its own remove button, and a sixth is refused with a short message. All the images go to the agent together in a single message, and the conversation shows them in one bubble.

2026.10.09 Files > Folder And Project Icons Use One Colour

Folder and project icons in Files, Projects and folder cards now share one blue in place of a mix of yellow and bright blue, the Platform Tour's feature icons use one icon colour instead of a different colour per step, and the dashboard greeting no longer ends with an emoji.

2026.10.09 Mobile > Databases Fit A Phone Screen

On a phone, a database's page now reads top to bottom. The name has its own line with the description beneath it, the header and the tabs share one margin, Pivot and Export wrap onto a second toolbar line, the grid narrows its Name column so the next column's values are readable, and the Tables row fades at its edge while there are more tables to scroll to.

2026.10.09 Models > Switch A Muse Spark Conversation To An OpenAI Model

Hardened switching models part-way through a conversation. A conversation that has been answered by Muse Spark now carries on when you switch it to an OpenAI model, with the earlier answers kept in context.

2026.10.09 Databases > Row Dialogs Use Underlined Fields On One Even Rhythm

The New Row and Edit Row dialogs now use minimal underlined fields with floating labels, evenly spaced down the dialog, and an accent line on the field you are editing. A reference field shows its values as small grey chips with a compact Choose or Edit action beside them, and keeps its height when empty. Star ratings sit on the same line style as other fields, the small action buttons are plain icons, and the Edit row tag sits beside the row's name.

2026.10.09 Agents > Busy Background Jobs Keep Running While They Call Tools

Hardened long-running agent jobs. A background job that spends several minutes calling tools, such as an agent working through a set of files, now counts as active for as long as its tools keep finishing, so it runs to completion. A job that genuinely stops responding is still ended as before.

2026.10.09 Mobile > Tapping An Agent On A Phone Opens Its Chat Every Time

Hardened agent taps on phones. Tapping an agent in the menu now waits for its chat to open, closes the menu and checks that both happened, retrying once and showing a short message if the chat still cannot open, and a second tap while one is in progress never opens a second chat. Each agent row on a phone also has a ⋮ button for Add to Favorites and Remove from Favorites, in place of the press-and-hold menu. Desktop is unchanged.

2026.10.09 Voice > Voice Conversations On iPad Keep Listening After The Greeting

Hardened voice conversations on iPad and other browsers built on Safari's engine. After the agent's greeting, Sapience now checks that your microphone is live and re-opens it if it has gone quiet, and if it still cannot hear you it shows a tap-to-retry button instead of waiting in silence.

2026.10.09 Settings > Daily Scheduled Jobs Show Their Time In Their Own Time Zone

Creating or editing a daily scheduled job, in Settings or on a project's Advanced tab, now saves the time you pick together with your time zone, so the job keeps running at that local time when the clocks change for daylight saving. Each job shows its time in its own zone, and the edit dialog names that zone so nobody in another zone mistakes it for their local time.

2026.10.09 Files > CSV Files With A Byte-Order Mark Download And Open Reliably

Hardened downloads of CSV files that agents create with a byte-order mark at the start. Their stored size now matches their contents, so they open from the file card and download in full, both when you are signed in and through a shared link.

2026.10.09 Databases > Add Column Guides You Through Each Column Type

The Add Column dialog now takes a column in order: its name, then its type, then that type's settings, with help text under each. A number column with missing settings shows a plain message saying what is needed, and on a database you can only read, the New view button is disabled and says why.

2026.10.09 Databases > Clearer New Row And Edit Row Dialogs

The New Row and Edit Row dialogs now give every field a clear label and value, use one consistent control for numbers, and separate Cancel and Save from the fields with a full-width rule. A currency field shows its currency sign even before you type.

2026.10.09 Notes > Discarding A Note Edit Removes The Comments You Added

A comment on selected text in a note now follows the note's own save. If you discard your edits, the comments you added since the last save are removed with them, so none are left behind without their highlight. Comments you added before pressing Save are kept, even if you close the note while it is still saving.

2026.10.09 Databases > Filter A View By A Row's Name

Saved views and filters can now match on a row's Name, the title shown in the first column, using contains, equals, is empty and the other text comparisons. Pick Name in the condition list when you create or edit a view, and the summary under the view reads the same way, such as Name contains acme. Through the API, the condition uses the reserved column key name in database view filters and in inline filters on /api/v3/databases/query.

2026.10.09 Databases > Forms Can Ask For The New Row's Name

A form built on a database table can now include a Name field, with its own label and help text and an optional required flag. The answer becomes the name of the row the response creates, instead of the form title and submission time, which is still used when the field is left out or left blank.

2026.10.09 Databases > Copy A Database's Structure Without Its Records

Copying a database now offers a Copy records too option, ticked by default. Untick it to get a fresh copy with the same tables, columns, views and forms but no rows, ready to reuse as a template. Developers can do the same by sending include_records: false to the copy endpoint (client.v2.srm.copy).

2026.10.09 Databases > See And Manage Who A Database Is Shared With

The Share dialog for a database now lists everyone you have shared it with, by name with their username and expiry date beneath. You can change a person's access between read and write, or remove it after a short confirmation, without leaving the dialog. A share created without an expiry lasts 365 days, and the dialog now says so.

2026.10.09 Databases > Person Columns Pick From Your Organisation

Editing a Person column in a row now offers a searchable list of the people in your organisation. Each person shows their name with their email beneath, so two people with the same name are easy to tell apart, and the grid shows the person's name rather than their username.

2026.10.09 Databases > Clearer Tables And Views Labels

On a database's page, Tables and Views now sit as small labels above their controls, so the tab row starts flush with the left edge under Tables and the views dropdown, which shows only the open view's name, is clearly marked as the place to switch views.

2026.10.08 Databases > Clearer Dialogs, Cells And Empty States

Databases get a polish pass. Dialogs open clean, use plain words and ask before discarding unsaved changes, cells honor their column types such as currency, stars, checkbox, link and date picker, and the row editor shows names instead of IDs and lists every invalid field at once. A new database and an empty table show guidance with Add first row and Add a column, copying a database names the copy and opens it, and deleting one asks you to type its name.

2026.10.08 Agents > Designer Agent Now Runs On Claude Opus 5.5

The Sapience Designer/Creative Agent now answers with Claude Opus 5.5 where your account is licensed for it, and with Claude Sonnet 5.5 otherwise. Before, it ran on Sonnet 5.5 for everyone.

2026.10.08 Models > Claude Opus 5.5 Holds More Before Compacting

Claude Opus 5.5 conversations now keep up to 400,000 tokens of history before Sapience compacts them, double the earlier 200,000. Long working sessions on Opus 5.5 keep more of their history word for word.

2026.10.08 Files > Long Recordings Are Transcribed In Full

Hardened transcription of long audio and video recordings. Recordings are now processed in shorter pieces so long recordings are transcribed in full, and a clear error is returned instead of a shortened text if a piece still cannot be completed.

2026.10.05 Chat > Sapience Suggests Separates Skills From Agents With Clickable Cards

The window that offers skills or a specialist agent before you send now opens under a Sapience Suggests: heading and shows the two kinds of suggestion in separate panels: skills in blue with square icons, agents in amber with round icons. Each suggestion is a card with a Use skill or Switch agent button, so it is clear what you can click, and the whole card picks it. Each panel has its own way to decline, Send without a skill or Keep the Sapience Agent, and Back to editing sits at the bottom. Hovering a skill or an agent shows its description in a tooltip, with agent descriptions in your language, and every card can be reached and picked from the keyboard. On a phone the button moves under the name so names and descriptions stay readable.

2026.10.05 Agents > Agents Are Marked In Amber Across Sapience

The colour that marks an agent is now amber everywhere it appears: the favourites board, the Sapience Suggests window, the agent lane beside long conversations and the agent picker for multiplayer chats. Purple has been retired from the whole app, so the badges on enterprise models and connected apps, the Tasks tile on a project's overview, tags, avatar colours and the colours used to tell item types apart now use neutral or warm tones instead.

2026.10.08 Models > Best-Model Aliases Keep Agents On Each Vendor's Latest Model

An agent's model can now be set to a best-model alias instead of a fixed model: anthropic-best-reasoner, openai-best-workhorse, google-best-cheap and so on, covering Anthropic, OpenAI, Google, Meta and xAI at three tiers (reasoner, workhorse and cheap). Sapience resolves the alias to that vendor's current recommended model every time the agent loads, so when a newer model arrives, every agent on the alias moves to it automatically with no change to the agent itself. The agent API returns the alias in model and the model actually in use in the new read-only resolved_model field, and usage and cost are always recorded against the real model. The Sapience Agent (Muse Spark) in the AI Store now runs on meta-best-reasoner.

2026.10.08 Models > Claude Haiku 5.5 Is Available

Claude Haiku 5.5, Anthropic's fastest and lowest-cost model, can now be chosen as an agent's model (anthropic/claude-haiku-5-5). It has a one-million-token context window with adaptive thinking, and it is the model behind the anthropic-best-cheap alias.

2026.10.08 Agents > Faster Perplexity Research And Deep Research Reports

The perplexity_web_research and perplexity_deep_research tools now run on Perplexity's Agent API, the same engine as the Sapience Web Research Agent. Everyday research answers come back in about two seconds, and deep research reports arrive in about half a minute rather than several minutes, still with clickable, titled sources. Agents set to an older Perplexity model such as Sonar Pro keep working and now answer on Sonar.

2026.10.07 Agents > Sapience Web Research Agent Can Work With Your Files

The Perplexity-powered Sapience Web Research Agent can now save, edit, move, copy, list, search, read and publish your files, create folders and email you what it found, so research lands straight in your Sapience files. It also gains the Google version's Coding, File Upload, Computer Vision and Image Generation abilities, so you can hand it images and documents to research from.

2026.10.09 Files > Pictures In An Office File With One Damaged Part Are Still Described

When a PowerPoint or other Office file contained a single damaged internal part, Sapience could not describe any of the pictures in it and the file's processing ended in an error. Sapience now skips just the damaged part and describes every picture it can read. A file that is damaged throughout is still reported as unreadable.

2026.10.09 Projects > Signing Out During A Shared Projects Refresh Stays Quiet

Signing out while Sapience was still refreshing your shared projects in the background could raise a needless error. Sapience now recognises that nobody is signed in and skips that refresh quietly.

2026.10.08 Agents > Sales Development Rep Is Now The Sapience Sales Prospecting Agent (SDR/BDM)

The Sales Development Rep (BDM) agent in the AI Store is now called the Sapience Sales Prospecting Agent (SDR/BDM). It also has powerful new tools for enriching people and companies that answer fast, and it reaches for them first on every lookup, keeping its earlier enrichment source as a backup for when they find nothing. It tells you which source each result came from. If you already use the agent, it keeps working as before under its new name, and agents that hand prospecting work to it reach it as usual.

2026.10.08 Chat > Conversation Navigation Shows The Day Of Each Message

The Conversation Navigation popup, opened by hovering the markers beside a long conversation, now groups its entries under the day they were sent, labelled Today, Yesterday, or the weekday and date. A Jump to date picker at the top takes you to the first message of any day in the conversation, exactly as if you had clicked that entry. The picker stays open while you choose, and the labels are translated into every language Sapience supports.

2026.10.08 Chat > A Message With An Unreadable Image Still Gets An Answer

When a message included an image Sapience could not open, such as a damaged file or a format it cannot decode, the whole turn stopped with an error. The agent now answers the rest of your message and tells you that the attached image could not be read, so you can send it again in another format. Other images in the same message are still used.

2026.10.08 Files > Steadier Handling Of Very Large Files

Hardened how Sapience reads very large files, such as spreadsheets of tens of megabytes, for better responsiveness under heavy use. The first time such a file is opened its text is shown as usual. If its text cannot be stored for reuse, the next request for it within the hour explains that clearly instead of reading the whole file again, and asking to refresh always tries again. Very large files are now read a few at a time, so a burst of them waits its turn instead of slowing everything else down.

2026.10.08 Agents > Schedules Set By An Agent Keep Their Local Time Across Daylight Saving

When you ask an agent to run something every day at a set time, the scheduled job now remembers that time in your own time zone, so a 7:00 am job keeps running at 7:00 am when the clocks change for daylight saving. On the day the clocks go forward, a time that does not exist runs at the first valid minute after it, and on the day they go back, a time that happens twice runs once. Through the API, a daily job accepts a time_zone together with schedule_hour_local and schedule_minute_local, and jobs without a zone behave exactly as before.

2026.10.08 Skills > PowerPoint Files Skill For Decks You Already Have

A new global skill, PowerPoint Files, guides an agent whenever the deck is one you supplied: reading it, changing it, or building a new deck on your own template. Questions about what a deck says are answered from the text Sapience extracted when you uploaded it, so they come back quickly without opening a workspace. Changes to the file itself, such as retitling a slide or adding speaker notes, run in run_in_sandbox and keep the deck's existing layouts, fonts and formatting. Every revision comes back as a new file in your Files, attached to the conversation, so your original is never overwritten. Slide Output stays the skill for a new deck built from your words; a request that gives it a deck or template to keep or change now goes to PowerPoint Files. Find it in the skill picker and the AI Catalogue under Slides/Presentations.

2026.10.08 Skills > Word Documents Skill With Real Tracked Changes

A new global skill, Word Documents, guides an agent whenever a Word file is the thing being read or changed. Questions about what a document says are answered from its extracted text. When you ask for changes to your own document, the agent edits it keeping its formatting, can redline it with real tracked changes and comments that you accept, reject or reply to in Word, and can fill in a template you supply. It also builds new documents that need real Word structure, such as headers and footers, page numbers, a table of contents or sections. Every revision comes back as a new file in your Files, attached to the conversation, so your original is never overwritten. A simple new document written from your words still uses save_file as before. Find it in the skill picker and the AI Catalogue under Writing & Communication.

2026.10.08 Agents > Agent Workspaces Now Carry The Word Skill Too

The workspace agents use through run_in_sandbox now also carries the latest Word skill, alongside the Excel and PowerPoint skills, and checks it for newer versions every night in the same way, so Word documents built or edited there follow the newest formatting and checking rules.

2026.10.07 Agents > Agent Jobs Run To Completion During Updates

Hardened long-running agent jobs so they run to completion when Sapience is updated, instead of being interrupted while they are still completing a background job.

2026.10.07 Menu > Announcement Banner Hides When You Change View

The full-width announcement banner now hides as soon as you move to another view in the app, not only when its five-minute timer ends. It still shows when you sign in and comes back on the next app load, and only its close button hides it for good.

2026.10.07 Chat > Long Conversation Compacting Recovers From A Stalled Step

Hardened compacting of long conversations. Each summarising attempt is now limited to 90 seconds and retried once, so a long conversation recovers instead of waiting up to ten minutes on a stalled step.

2026.10.07 Sharing > Sharing A Container Shares Everything Inside It

Sharing a project, folder, page, goal or database with someone now always shares everything inside it, at any depth, including sub-folders and files homed beneath it. The app sends complete shares by default, and the API refuses a share on one of these items that leaves out what is inside it, with a clear message.

2026.10.07 Settings > Customize Dashboard Moves Into Quick Links

The dashboard header is back to one line. Customize Dashboard moved out of a second row under the Dashboard and Home toggle and into Quick Links, in the place Recent Activity used to hold, styled like the other quick links. The Home screen is unchanged.

2026.10.07 Databases > All Databases Page And Project-Style Database Header

Databases now has an All Databases page that shows your databases as cards or as a list, with search, a filter for Active, Archived, Templates, Shared, Org or All, and a New Database button. Opening a database shows a Projects-style header with its name, description, a favorite star, refresh, a green Chat button and an Actions menu for editing the description, renaming, tags, sharing, moving, duplicating, adding a table or form and deleting.

2026.10.07 Models > Muse Spark Falls Back Quickly When Overloaded

Hardened Muse Spark 1.3 conversations. When the service is overloaded, a conversation now switches at once to Muse Spark 1.2 and then to Claude Sonnet 5.5, instead of retrying for about five minutes before failing. Other models' fallback behavior is unchanged.

2026.10.07 Projects > New Folder Button Removed From Projects Page

The New Folder button is gone from the Projects page, in both card and list view, because a project is always a top-level item. Folders can still be created inside a project's own tabs.

2026.10.07 Voice > Push-To-Talk Starts Only On A Solo Key Hold

Push-to-talk now starts only when you hold its key on its own for a moment, and pressing any other key with it cancels, so Cmd+V no longer starts a dictation on a Mac. In Settings, Advanced the key list is replaced by a Change PTT Key button: click it, then tap the key you want, with Esc to cancel and Backspace or Delete to turn it off.

2026.10.07 Accounts > An Expired Login Signs You Back In Or Says Why

When your login expires, the desktop app now signs you back in silently with your saved sign-in details and retries the download or preview you asked for, so it simply completes. On the web, or when the desktop sign-in can't complete, Sapience signs you out cleanly and the sign-in page tells you your login has expired and to sign in again, instead of showing an error such as "Download failed". This applies to every part of the app, including a login that reaches its expiry time while the app is open, and the message appears in every language Sapience supports.

2026.10.07 Mobile > Tapping An Agent Shows Progress Straight Away

On phones, tapping an agent in the menu now shows the loading indicator above the menu while the agent's chat opens, so a slower start never looks like nothing happened. From any screen other than the home screen, the menu also closes as soon as you tap.

2026.10.07 Models > Muse Spark Conversations Recover From Expired Reasoning Context

Hardened Muse Spark conversations for long-running and resumed chats. When the model reports that a conversation's saved reasoning context has expired, Sapience now clears that context from the conversation and retries, so the turn answers and later messages continue normally. Your messages and the agent's replies are always kept.

2026.10.07 Models > Claude Turns With Skills And Attached Files

Hardened how app context, such as an armed skill or a note about an attached file, is placed in requests to Claude models, so turns that combine them complete reliably. The context still reaches the model in full, carried inside your message.

2026.10.07 Chat > Conversations Open With An Introduction To The Agent

The top of every conversation, above the welcome message, now introduces the agent you are talking to: a card with its name, its full description, and an About this agent link for agents that have an About page. The card always shows the whole description, so you can read what an agent is for without opening anything. It does not appear in group rooms or for agents that have no description.

2026.10.07 Chat > A More Compact Chat Header

The chat header takes less room. The agent's name now sits under a small Agent label, and the description no longer fills lines in the header: hover over the agent's name to see the full description in a card, or click it to open the About window. The Focus Mode link is gone, since the header no longer has a description to hide. Add People and Add Agent are now one Add menu, with a short explanation of each choice.

2026.10.07 Chat > Welcome Messages Are Labelled

An agent's welcome message now opens with a bold Agent Welcome Message line, in every language Sapience supports, so the agent's standard greeting is easy to tell apart from the conversation that follows.

2026.10.07 Notes > Deleting A Comment Removes Its Highlight Straight Away

Deleting a comment that was posted on selected note text left its yellow highlight in the note, and hovering the text kept showing the comment until a refresh. With the note open in the editor, deleting the comment now removes it from every comment list and the comment count, stops it appearing on hover, and takes the highlight off the text. The note is saved at once, so the change holds after a refresh, a Discard or an undo.

2026.10.07 Email > Deep Research Emails Show The Question You Asked

The "Your Deep Research" email could show text the agent adds around your question, rather than the question itself, in its subject line and its "Your Query" line. Both now show your question exactly as you typed it, for finished and for failed research.

2026.10.07 Agents > Agents Can Read The Current Conversation Word For Word

get_conversation_transcript can now read the conversation the agent is working in. Called with no conversation_uid, it returns the full transcript of the current conversation, so you can ask an agent to go back to your exact wording, for example after earlier turns of a long conversation have been summarized. Reading a past conversation works as before: the agent finds it with search_conversation_memory and passes the conversation_uid it returns.

2026.10.07 Agents > find_tool Marks Tools That Are Already Loaded

When an agent that looks up its tools as it needs them calls find_tool, any matching tool that is already loaded is now named on its own line as ready to call. This covers tools loaded in advance by the Jev shortlist and tools loaded by an earlier search in the same conversation. Previously only the tools a search had just loaded were named as callable, so an agent could treat a tool it already had as one it still needed to load. Tools that a search loads are listed as callable, as before.

2026.10.07 Sharing > One Clear Message When A Project Share Is Refused

Sharing a project from its Team tab with a username that has no Sapience account showed two red error messages for the same refusal. It now shows one message that names the person and the reason. Shares with people in your organization already worked this way and are unchanged.

2026.10.07 Files > File Types That Can't Be Searched Are Set Aside Straight Away

Hardened search indexing for files whose type can't be indexed, such as .eps design files or .opj project files. They are now set aside as not searchable the first time they are processed, rather than being retried in the background. Files that couldn't be read for a temporary reason, such as audio or video while transcription is briefly unavailable, are still retried and indexed once it recovers. Every other file type is unaffected.

2026.10.06 Menu > Actions Menus Open From The Whole Button

The Actions dropdown in the chat header is now medium size, matching the project and note headers, and is as wide as its label instead of a fixed width. Clicking anywhere on the button, including the word Actions, now opens the menu everywhere the Actions split button appears.

2026.10.06 Connected Apps > MCP Servers Can Load Only On Chosen Agents

An MCP server can now be set to load only on the agents it is attached to, instead of on every chat in its scope; this works for user, organization and system servers. Organization admins can manage organization MCP servers and attach or detach them to agents. MCP servers that use standard OAuth sign-in can now be connected through that sign-in flow, and when a stored login expires the agent is told to reconnect. In the API, GET /api/v2/mcp-servers accepts agent_uid and servers carry a load_mode field.

2026.10.06 Outbound > Split List Among Senders Campaign Setting

Each outbound campaign now has a Split list among senders checkbox on its Sending & schedule tab. On by default it behaves as before; switched off, the owner's seat keeps its usual path and every other paired LinkedIn seat also sends its own invite-only request to the same leads. Re-invite protection is applied per seat.

2026.10.06 Databases > Chat With A Database From Its Screen

The Databases screen now has a green Chat button beside a database's title. It opens a floating chat, which can go full screen, with that database's own agent, and clicking it again focuses the chat that is already open. People with read-only access can chat too, and the database agent can look up and use tools beyond its fifteen built-in database tools.

2026.10.06 Databases > Add A Markdown Description To A Database

A database can now have a description written in Markdown. It shows under the database's name on the Databases screen, clamped to three lines with the full text on hover, and hovering a database in the left bar or in the Everything view shows the whole description. Owners and people with write access can add, change or clear it from a Description button with a live preview, and read-only viewers cannot edit it.

2026.10.06 Agents > Designer/Creative Agent Renamed And Described By Capability

The global design agent is now called Sapience Designer/Creative Agent; it was Sapience Marketing Agent (Design, Creative). Its description, welcome message and store listing now describe what it can do without naming a model or vendor.

2026.10.06 Chat > Sapience Suggests Offers To Plan Big Requests First

With the Preview suggestions setting in Settings, Advanced turned on, a big multi-step request now gets a Plan this first option in the Sapience Suggests window. Picking it turns on plan mode for that one message only, and Send without a plan sends as normal. It is never offered when plan mode is already on.

2026.10.06 Projects > Link Related Projects

You can now mark two projects as related. Open a project's Actions menu and choose Link project to another, then pick the other project; both projects now list each other. A new Related projects card on the project's Overview tab, in the project page and the project pop-up, shows every linked project: click a name to open it, or press Unlink to remove the link from both projects. A linked project you can no longer open appears as Unavailable project and can still be unlinked.

2026.10.06 API > Project Relationships Stay Consistent And Appear In Project Lists

Hardened how project relationships are stored. Updating a project with PUT /api/v2/sapience/{uid} keeps its existing relationships when the body does not include the relationships field, and replaces them only when you send it. Linking and unlinking at the same moment from different places no longer drops a link. Duplicating a project no longer copies links to projects outside the copy, and deleting a project removes its links from the projects it was linked to. If the reverse side of a two-way link cannot be written, the add-relationship response message now says so. Project rows from POST /v2/srm/tree with projection meta and include_counts now carry each project's relationships, and the TypeScript SDK type SapienceObjectMeta includes the new optional relationships field.

2026.10.06 Projects > Spreadsheets Added To A Project Go To The Project Agent's Data Analysis

Adding a csv, tsv or xlsx file to a project, by uploading it into the project or any folder in it or by regenerating the project's Knowledge Library, now gives the project agent that file for data analysis with Code Interpreter instead of indexing it for search. PDFs, documents and every other file type still go to the project's searchable Knowledge Library as before. Removing a spreadsheet from the project removes it from the agent's analysis files automatically, and there are no new questions or settings to choose.

2026.10.06 Outbound > Sync Button In The Inbox

The Inbox in the Outbound Ops Console now has a Sync button that runs the same LinkedIn inbox sweep as the hourly job, whenever you ask. It shows progress while it works, then reloads the thread list and the counts. A reply it finds from a lead in a campaign stops that lead's sequence, exactly as the hourly sweep does. Sync only reads conversations and never sends anything, and clicking it while a sweep is running shows that sweep instead of starting another.

2026.10.06 Outbound > Launch Shows How A Sales Navigator List Import Ended

When you pick a list in Launch, the Outbound Ops Console now shows how its Sales Navigator import ended: failed, with the provider's reason; still running, with members to follow; or finished with no people, with the reason, such as the spend limit. Previously a stalled or failed import simply showed 0 people. Retrying the same search into a second list keeps each list's result separate, and import tracking is hardened to follow long-running imports through to completion.

2026.10.06 Outbound > Configure LinkedIn Sending For A Draft Campaign

Under Sending seats, a draft campaign with paired seats now offers Configure LinkedIn sending, which adds the paired seat to Sending seats with the usual Arm, Pause and Resume controls. Configuring never sends anything by itself, and it is refused for a campaign that has ended. A campaign that the sending schedule does not include now shows a Not scheduled badge with a hint, so the console never suggests it is sending when it is not.

2026.10.06 Agents > Sapience Web Research Agent Now Runs On Perplexity, With A Google Version Alongside

The Sapience Web Research Agent now answers with Perplexity's Sonar Pro, which returns cited answers noticeably faster for everyday lookups and keeps the clickable source links on every claim. It stays featured in the AI Store under Research. The Google Search version is still available as Sapience Web Research Agent (Google), for when you specifically want results grounded in Google Search. Agents that look something up on the web during a conversation now use Perplexity by default, and a web search started from the chat bar or Quick Chat goes straight to the Sapience Web Research Agent.

2026.10.06 Notes > Printing A Note Fills The Page From The Full-Page Viewer And The Editor

Printing a note from the full-page note viewer now starts at the normal page margin and uses the full width of the sheet, instead of leaving the left half blank. Printing from the note editor now keeps the menus, formatting toolbar and buttons off the paper and fits the note inside the printable area across as many pages as it needs. Printing from the pop-up viewer is unchanged.

2026.10.06 Databases > Redesigned Row Editor, Two-Line Text Cells And Removable Files Before Adding

The Add Files window is wider and lists each file on its own row with a Remove action, the same pattern as an agent's Knowledge Base, so you can drop one file before adding the rest. Long text cells show one line by default; one click reveals a second line and a double-click opens a scrollable editor for the cell. The expanded row view now uses bordered fields with a clear focus state, plus and minus steppers for numbers, a resizable box for text, linked values as chips with a Linked from label and a Choose reference rows button, and files as cards showing their type with a remove button. Long values wrap instead of being cut off.

2026.10.06 Files > Files Added On Another Device Appear Without Refreshing

When you upload a file, or a file you can open is added to a project, your other open Sapience windows and devices now receive it straight away and show it in the file list and the project's Files tab without a refresh. Only people who can open that file are told about it.

2026.10.06 Files > More Reliable Downloads In Long-Running Sessions

File downloads and spreadsheet previews now always use your current sign-in, improving reliability in sessions that stay open for a long time, such as the desktop app. If a download still cannot complete, the message now says what to do, for example to try again after your session refreshes, rather than showing a bare error code.

2026.10.06 Sharing > Agent Public Links Need A Power Account

Previously, anyone who owned an agent could create a public link for it, whatever their account level. Publishing an agent is now a power-tier capability: an account below that tier which tries is told to upgrade or to talk to its org admin. Power users, org admins and all existing links are unaffected.

2026.10.06 Agents > Sapience Agent Creates A New Database From A Description

Ask the Sapience Agent for a database in plain words, such as "create a new database that will hold data from conference survey participants", and it designs the tables and typed columns and creates the database for you. Follow-up requests in the same conversation, such as adding a participant, are handed to that database's own agent, which writes the rows.

2026.10.06 Databases > Database Agent Lists Who Has Access And Publishes Public Forms

When you chat with a database, its agent can now tell you who has access to it: the owner, whether your whole organisation can read it, each person with read or write access and where that access comes from, and any public form links. You can also ask it to create a form on one of the database's tables and publish it as a public link, optionally protected by a password, and it gives you the link to share. Creating a form needs edit access to the database; with read-only access the agent tells you so and creates nothing. The database agent now also reasons at a higher effort level by default.

2026.10.06 Accounts > Log Out Turns Off Notifications Before Ending Your Session

When you log out, the app turns off notifications for that browser and saves that setting. That step used to run after your session had already ended, so it could fail quietly. It now finishes before your session ends. If the notifications step takes more than a few seconds, logging out goes ahead anyway, so you are never left signed in.

2026.10.05 Menu > Shared Conversations Group Is Renamed Shared

In the sidebar's Conversations list, the group that was labelled Shared With Me is now labelled Shared, because it also holds conversations you have shared with other people. The new label is translated in every supported language.

2026.10.05 Databases > Forms Follow Their Builder's Time Zone And Stay Unlisted

A form now stamps each response with the time zone of the person who built it: the generated row name and the owner's notification email show that local time instead of Sydney time. Public form pages ask search engines not to list them, and star-rating columns are capped at 10 stars, both when a column is added and when a table is created with columns.

2026.10.05 Databases > Databases Sidebar Shows Bold Names And Kind Chips

In the Databases sidebar, database names are now bold and each database carries a Database chip pinned to the right, like the Table and View chips. All sidebar chips are smaller with a fill, and the expand arrow moved to the left so the right edge is free for the chips.

2026.10.05 Chat > Sapience Suggests Shows Formatted Descriptions In A Larger Window

The Sapience Suggests window that offers a skill or a specialist agent before you send now shows each description formatted, so bold text and emphasis appear as intended instead of as raw markdown symbols. Descriptions run to three lines instead of two, and the window is wider and a little taller so more of each description fits beside its action button. On a phone it still fits the screen. Links inside a description appear as plain text, so clicking a card always just picks it, and the hover tooltip shows the same text without markdown symbols.

2026.10.05 Chat > A Pasted Image Clears From The Message Box When You Send

After you pasted an image and sent your message, the image preview stayed above the message box until the agent finished replying. It now clears the moment you press Enter, while the agent still receives the image with that message. If the send fails, the image returns to the message box so you can try again, and a new image you pasted in the meantime is never replaced. Messages the app sends on your behalf afterwards, such as queued messages, never pick up an image left in the box.

2026.10.05 Notes > Printing A Note Prints The Whole Note

Printing a note from the note viewer with your browser's Print command printed only the part visible on screen. The whole note now flows across as many pages as it needs, from both the full-page viewer and the pop-up viewer. When the note is open in the pop-up, the page behind it stays off the paper, and the sidebar, headers, note toolbar and floating table of contents are left out of the printout.

2026.10.05 Meetings > Meetings Window No Longer Labelled Preview

The Meetings window's header now reads Sapience Meetings, without the Preview tag that remained from before Meetings became generally available.

2026.10.05 Projects > Add Menu Groups The Note Actions Under Notes

In a project's Files & Notes tab, the Add menu now keeps its two note actions together under a Notes sub-menu: Add a Note and Add Existing Note. Upload Files/Folders and New Folder stay at the top level of the menu, so the list is shorter and the note actions are easier to find. On a phone the same actions appear as a Notes group in the flat menu.

2026.10.05 Projects > Make A Copy Of An Existing Note Into A Project

The Add Existing Note dialog now offers Make a Copy into Project beside Move Into Project. Move takes the note out of its current location, as before. Make a Copy adds a duplicate to the project and leaves the original where it is, so a note can live in more than one place without being moved.

2026.10.05 Skills > Excel And Spreadsheets Skill For Workbooks With Live Formulas

A new global skill, Excel & Spreadsheets, guides an agent whenever a spreadsheet file is the thing being read or produced: building a tracker, budget, forecast or fill-in template, changing a workbook you already have, or turning a messy CSV, TSV or xlsx export into a proper workbook. The agent builds the file with run_in_sandbox, using live formulas instead of pasted numbers, keeping inputs in labelled cells, and using the exact tab names, headers and formulas you asked for. When it edits your own file it matches the existing conventions and writes only where the task needs it. Every formula is recalculated and checked for errors before the file comes back, and key results are spot-checked against the inputs. The finished workbook is saved to your Files and attached to the conversation, and the agent tells you the headline numbers, every assumption and the cell where it lives, any cells you still need to fill in, and anything it could not check. Questions answered from a sheet are handled with run_python and answered in the chat, and valuations, charts, dashboards and narratives go to the Financial Models, Charts, Data Dashboard and Data Storytelling skills. Find it in the skill picker and the AI Catalogue under Data & Analytics.

2026.10.05 Agents > Agent Workspaces Keep Their Excel And PowerPoint Skills Current

The workspace agents use through run_in_sandbox now carries the latest Excel and PowerPoint skills and checks for newer versions every night. When either skill changes, the workspace is rebuilt with it, so workbooks and decks built there follow the newest formatting and checking rules without waiting for a release.

2026.10.05 Skills > Image Generator Skill For Better AI Images

A new global skill, Image Generator, guides an agent whenever it creates an image with generate_image. It first checks that an image is the right answer, and hands charts, diagrams and anything needing exact text or exact brand colours to the Charts, Mermaid Diagram, Inline SVG Graphics or Infographic skills instead. It picks the size from how the image will be used (landscape for banners and heroes, portrait for posters, square otherwise) and the quality from the stage (low to explore, medium by default, high for a final). It writes a structured prompt covering subject, setting, style, composition, light, colour and mood, keeps lettering in the image short, uses your organization's style guide colours for branded work, and leaves clean space for a logo rather than trying to draw one. It is clear that it creates new images from a description and cannot edit an existing photo. Find it in the skill picker under Slides & Visuals, or in the AI Catalogue under visualization.

2026.10.05 API > Record Jev Decisions With The Report Event Endpoint

POST /api/v2/system/report-event now accepts a second event type, frontend_jev_decision_made, so a client can record a decision Jev made, such as which skill or which reply mode it chose. The body has three fields: question_type, a label of up to 100 characters such as which_skill; answer, any JSON value except null; and additional_info, an optional JSON object for anything else worth keeping. Any other field is rejected, including user_id and org_id, because the event is always recorded for the signed-in user and their organization. Each call records one event, so repeating a call records another. In the TypeScript SDK it is client.api.v2.system.reportEvent; send event_type even though the generated type shows it as optional.

2026.10.05 Chat > Sapience Agent Suggests A Better-Fitting Agent Before You Send

With Settings ▸ Advanced ▸ Preview: Jev skill, mode & agent suggestions turned on, a message you send to the Sapience Agent is now also checked against every agent you can use, installed or not, using each agent's name, description and welcome message. If one fits your request clearly better, it appears in the same popup as the suggested skills, with Keep the Sapience Agent as the other choice. Picking an agent switches who replies, just like @-mentioning it: the conversation stays the same, and your message goes to that agent. A casual message offers nothing, and the check runs alongside the skill suggestions, so it adds no extra wait. If the check is unavailable, your message sends exactly as it would have.

2026.10.05 Chat > Reply Mode Chosen For Each Message, Shown On A New Chip

With the same preview setting on, each message you send is also given the reply mode that suits it: Minimal-Fast for a quick factual question, Think Longer for a request that needs careful analysis, and Balanced otherwise, which is kept unless another mode is a clear fit. A new chip beside the message box always shows the mode in use, and clicking it opens the mode picker. A mode you choose yourself, from the picker or from a link that sets one, is never changed for the rest of that conversation.

2026.10.05 Chat > Messages Sent After Your Sign-In Expires Go Through On Desktop And Mobile

Sending a chat message after your sign-in had expired could fail with a generic error, even where the app could renew your sign-in on its own. In the desktop and mobile apps, sending a message now quietly renews your sign-in and sends the message once more, so it goes through without you having to retry. On the web, an expired sign-in still shows the same short failure line in the chat.

2026.10.04 Skills > An Armed Skill Applies To One Message Only

A skill you arm for a message, by picking it in the skill dropdown, opening a link that arms it or accepting a suggestion, now applies to that one message only, as a skill typed with a slash already did. Before, it stayed armed for every later message in the conversation. If the send is blocked or you go back to editing, the skill stays armed.

2026.10.04 Chat > Skill Suggestions Keep Working After A Reload

Hardened skill suggestions in Settings, Advanced, Preview: with suggestions turned on, they now keep working after you reload the page or reopen the app and your session is restored.

2026.10.04 Chat > Simple Questions Get Short Direct Answers

A simple one-fact question now gets one or two plain sentences, without headings or suggested follow-ups, while substantive requests keep the full layout and suggestions. Agents that look up their tools on demand now begin replying sooner.

2026.10.04 API > Webhook Event Names Made Consistent

The public webhook event names are now user_created, database_row_added, database_row_updated and database_row_entered_view; the other seven names are unchanged. Creating a webhook still accepts the four earlier names and stores the new ones, existing subscriptions under the old names keep firing, and each delivery's type and X-Sapience-Event header now carry the new name.

2026.10.04 Menu > Favorites Conversations Group Starts Collapsed

In the sidebar's Conversations list, the Favorites group now starts collapsed, like Scheduled Jobs, Team Chats and Shared. Only Recent starts open, so the list opens short and you expand a group when you want it.

2026.10.04 Agents > Short Conversations With Very Large Items Stay Within The Model Limit

Hardened how an agent fits a conversation into its model's limit when a few items are very large, such as the contents of attached files or a big tool result. After the usual compaction, if the request still doesn't fit, Sapience now shortens the largest earlier items, keeping their start and end, with a note saying how much was removed and that specific sections can be read again. It shortens only as much as needed, largest first, and leaves your latest message untouched. If your latest message is too large for the model on its own, the agent stops with a message suggesting you shorten or split it.

2026.10.04 Databases > Forms That Turn Public Responses Into Table Rows

You can now build a form on any Databases table and share it as a public link, so people without a Sapience account can send you answers that arrive as new rows. Choose New form, pick the table and the fields to ask for, set each field's label, help text and whether it is required, and add a title, description, submit button text and thank-you message. Publish the link with an optional password; form links have no view limit. Visitors get a mobile-friendly page with the right input for each column type, including file uploads of up to 10 MB per file and 10 files per response. If an answer is missing or invalid, the form is shown again with the problem marked and nothing is saved. Each response becomes exactly one row, and a new Source column shows which form it came from; for developers, those rows carry system.source with kind form and the form's uid. You get an email for every response, which you can turn off per form, and uploaded files are only sent for AI processing if you switch that on for the form. You can stop accepting responses, reopen the form or revoke the link at any time, and your forms appear in a Forms group next to Views and Tables in the Everything view. Database owners and members who can edit the database can build forms; read-only members cannot.

2026.10.04 Skills > Sapience Skills Manager Groups Skills By Scope

The Sapience Skills manager used to list every skill in one long, newest-first table, so your own and your organization's skills were buried among the global ones. It now groups the list by where each skill comes from: your own skills first, then your organization's, then the global library, each in a collapsible group headed by its name and a count, such as User (9). The global group starts collapsed so the list is no longer overwhelming, and the groups you use most are open. Searching opens every group that holds a match and hides the ones that don't, a search with no match says so, and the select-all checkbox ticks only the rows you can see, showing a partial state when only some are ticked. Which groups are open carries over when you switch the list to fullscreen. The same change retires the purple on this screen: your own skills now carry a teal badge.

2026.10.04 Agents > Tool Search Shortlist Behaviour Updated

This entry updates the two 2026.10.03 entries on the jev shortlist preprocessor and on tool search shortlists with jev and llm, so they describe the final behaviour. Before the agent's first step, Jev is asked one question each about tools, agents and skills, and each question includes a "none of these fits" answer. If "none" is Jev's top answer for a kind, nothing of that kind is suggested. Otherwise up to three items Jev is reasonably confident about (a score of at least 0.20) are suggested in a private note the user doesn't see, and the suggested tools are loaded ready to call. When nothing fits, the agent gets no note and nothing extra is loaded. On our 196-request test set the right tool is among the suggestions 95.9% of the time (agents 100%, skills 100%), and on requests that need nothing a tool is loaded only 2.5% of the time. When one agent calls another, the called agent is not offered other agents, is not offered skills if the calling agent named one, and is always offered tools. There is no llm shortlist: the llm setting changes only how find_tool searches, and an agent on llm starts with no private note.

2026.10.03 Files > Files Page Groups Your Projects Under One Row

On the Files page the root now lists all your projects under one collapsed Projects row instead of scattering them among your files and folders. Expand it to see each project; opening, right-click menus and drag and drop work as before, and collapsing the row clears any selection inside it. The row uses the blue project folder, and the icon, grip and checkbox line up with each name in the tree view.

2026.10.03 Projects > Save Notes Offline With Files From Projects

Saving a mixed selection of files and notes from a project for offline use now saves the notes in full alongside the files, and ordinary file downloads keep working. If something cannot be saved, the summary names the reason, such as storage being full or unavailable, instead of reporting success.

2026.10.03 Agents > Plan Follow-Ups Work On Claude Models With Extended Thinking

When an agent on a Claude model with extended thinking finished a turn with plan steps still open, its automatic follow-up to close those steps could be turned away, leaving the plan unfinished. The follow-up now replays the earlier answer exactly as the model produced it, in its original order, so plan steps close as intended. The answer you see is unchanged, and other models are unaffected.

2026.10.03 Agents > Jev Shortlist Preprocessor For Tool Search

An agent that looks up its tools as it needs them can now choose how it searches. Set tool_search_strategy in the agent's mba_agent_config, next to tool_search_first, to bm25 (keyword search, the default), jev or llm. On jev, a jev shortlist preprocessor ranks every tool, agent and skill the user is allowed to use before the agent's first step. The top three tools, agents and skills it finds are listed in a private note the user doesn't see, the top tools are loaded ready to call, and an optional quality cutoff can drop weak matches so a request with no good match gets no note and no extra tools. A likely skill's instructions are also loaded in advance when Jev is confident. The agent still makes its own choice and can call find_tool for anything else. find_tool uses the same method the agent is set to: jev uses TypeSafe's Jev, a fast decision model, and llm uses Gemini 3.8 Flash. An agent's tool list now stays stable from one message to the next in a conversation, so the conversation stays cached. If search is unavailable, find_tool says so and looking a tool up by its exact name still works. Agents without the setting are unchanged.

2026.10.03 Agents > Effort Preprocessor Picks Reasoning Effort For Each Message

Before each message, an effort preprocessor now picks low, medium or high reasoning effort for that message, so a simple question is answered quickly and a hard one still gets deep thinking. It is on for every agent. An effort you set on the request, or one pinned by a skill, always wins, and agents handed work by another agent get their own pick. The choice takes a fraction of a second. If the preprocessor can't answer in time, the agent's configured effort is used.

2026.10.03 Agents > Agents Carry On Past Malformed Tool Calls

Hardened how agents handle a tool call that the model writes with arguments it can't read. The agent now carries on with the conversation, so a long, multi-step answer completes. This applies to every agent and every model.

2026.10.03 Projects > Everything View Adds Conversations, Favorites And Connected Apps

The Everything view now opens on eight bold top-level entries: Databases, Agents, Skills & Workflows, Projects, Files & Notes, Favorites, Connected Apps and Conversations. Every project sits under Projects, which uses the same blue project folder as the projects themselves, and Files & Notes lists your root folders, files and notes exactly as the Files page does, so the two screens always agree. Favorites lists your starred items. Conversations is organised like the sidebar, with Favorites, Scheduled Jobs, Team Chats, Shared With Me (when you have any) and Recent groups that you expand to see the conversations inside, and opening one takes you straight to it. Connected Apps is split into Individual and Team & Enterprise, with Google Apps and Microsoft Apps as expandable groups, and each app is labelled App in the Type column. Skills & Workflows now always shows three groups, My Skills, My Organization's Skills and Global Skills, which you expand to see the skills inside. The labels and error messages in the view are translated into all supported languages.

2026.10.03 Projects > Everything View Updates Without Pressing Refresh

Tasks, goals, notes and files you create elsewhere, from another tab, the project dashboard or an agent, now appear in an already-expanded project in the Everything view straight away, with the section counts kept in step. The duplicate Refresh and Clone buttons above the view have been removed, leaving the one Refresh next to the tree.

2026.10.03 API > Webhooks For Database Record Updates And Views

Webhooks can now fire when a record in a Databases table is updated (domain_databases_row_updated) and when a record enters a saved view (domain_databases_row_entered_view). Pass a database_trigger on /webhooks/create with the table, optional field conditions (contains, is, is empty and more) and, for the view event, the view. Conditions also work on the existing record-created event, and a record-updated webhook with conditions fires when the record starts to match them. Webhooks without a database_trigger behave as before. Column configuration for file columns can also be changed after creation.

2026.10.03 Projects > Databases Open-Row Button, Object Links And File Cells

In Databases, the open-row button is now always visible, so it works on phones and tablets without hovering. Linked notes, tasks and goals open from the row name just like files do, long text cells have an expand button with a copyable popover, and file cells use the same upload-or-select picker as the rest of the app with a popover to preview, download, rename or detach attached files. File columns have a Show or Hide summary option in the column menu, and a multi-file cell now shows one labelled summary per file. The tables bar, toolbar, search and grid now share one aligned, borderless layout, and a right-click Copy ID option copies the database, table, column, view or row ID.

2026.10.03 Agents > Tool Search Shortlists With jev And llm

An agent that looks up its tools as it needs them can now choose how it searches. Set tool_search_strategy in the agent's mba_agent_config, next to tool_search_first, to bm25 (keyword search, the default), jev or llm. On jev and llm, a shortlist preprocessor ranks every tool, agent and skill the user is allowed to use before the agent's first step: the top three tools are loaded ready to call, and a private note the user doesn't see lists the best-ranked tools, agents and skills. On jev, a likely skill's instructions are also loaded in advance when Jev is confident. The agent still makes its own choice and can call find_tool for anything else, and find_tool searches with the same method the agent is set to. jev uses TypeSafe's Jev, a fast decision model; llm uses Gemini 3.8 Flash. If search is unavailable, find_tool says so and looking a tool up by its exact name still works. Agents without the setting are unchanged.

2026.10.03 Agents > Effort Preprocessor Picks Reasoning Effort For Each Message

Before each message, an effort preprocessor now picks low, medium or high reasoning effort for that message, so a simple question is answered quickly and a hard one still gets deep thinking. An effort you set on the request, or one pinned by a skill, always wins, and agents handed work by another agent get their own pick. The choice takes about half a second. In our tests across simple, medium and hard prompts, the median time to a complete answer fell by about 40% in two separate runs, and by 40% to 57% on simple questions, with answer quality unchanged. If the preprocessor can't answer in time, the agent's configured effort is used.

2026.10.03 Agents > Agents Carry On Past Malformed Tool Calls

Hardened how agents handle a tool call that the model writes with arguments it can't read. The agent now carries on with the conversation, so a long, multi-step answer completes. This applies to every agent and every model.

2026.10.03 Agents > Agents Are Told When A File From Their Code Isn't Saved

When an agent's Python tool or coding sandbox writes a file that can't be saved, such as a file that came out empty, the agent is now told which file it was and why, so it can correct its code instead of running it again unchanged. Files that were saved are listed as before.

2026.10.02 Chat > One Actions Menu In The Chat Header

The chat header now has one Actions dropdown with two labelled sections: Agent Actions (Edit, Delete, Share Agent) and Conversation Actions (Rename, Info, Favorite, Share, View Files, PDF Export, Add People, Add Agent, Delete Conversation). The separate Actions link on the conversation banner is gone, leaving only the Add people and Add agent links. Share Agent opens the share dialog, and the Published row appears beneath once the agent is published.

2026.10.02 Databases > Databases Open Faster And Update Live

Opening a database now loads its tables, views and rows in one pass, and switching between tables and saved views is near-instant from what is already loaded. Rows, columns and structure that other people change in a shared database appear for you without refreshing.

2026.10.02 Models > Gemini 3.8 Flash Falls Back To Muse Spark 1.3

Hardened Gemini 3.8 Flash turns. When the model cannot answer, an agent now switches to Meta Muse Spark 1.3 for that turn instead of failing after its retries, and a short notice tells you the model was switched.

2026.10.02 Models > OpenAI GPT-6.1 Sol Is Available

OpenAI's GPT-6.1 Sol can be selected for chat and agents: it appears in the chat model picker, the create-agent wizard, Advanced options and the AI Store model browser, listed just below GPT-6 Astra. It accepts images, supports deep research and adaptive reasoning, has a 1,050,000-token context window and an adjustable reasoning effort (default medium), with no temperature setting. It also replaces GPT-6 Sol as the model GPT-6 Astra falls back to.

2026.10.02 Notes > Delete A Page Together With Its Files

The confirmation for deleting a page that holds files now names the page, shows one dialog at a time when several pages are deleted, and warns when the page also contains notes or folders that will be deleted too. If you are not the page owner the option to delete its files is not offered and you are told only the owner can; if other people's files or notes sit inside, a plain message says nothing was deleted. Cancel is quiet.

2026.10.02 Accounts > Scheduled Jobs Of Inactive Accounts Stop Running

Hardened scheduled jobs: a job owned by an account that is no longer active no longer runs, so it no longer starts agents, saves replies or sends notifications. Platform maintenance jobs are not affected.

2026.10.02 Chat > Created By, Type And Filters In All Conversations

The All Conversations list now has a Created By column showing who owns each conversation, a Manual or Scheduled badge on every row, and sorting on these columns, with newest-updated first still the default. Filter chips for Favorites, Scheduled Jobs, Team Chats, Shared With Me and Recent use the same grouping as the sidebar, so a conversation sits in the same group in both places.

2026.10.02 Projects > Scheduled Runs Grouped In A Project's Conversations

A project's Conversations tab now lists your own chats first and gathers scheduled-job runs into a collapsed Scheduled Jobs group, the same way the sidebar does, with the number of runs on the group header. Searching the tab also searches the scheduled runs and opens the group while a search is active.

2026.10.02 Notes > Callout Blocks And Boxed Code Blocks

Notes now have callout blocks. Choose Callout from the editor's Insert menu, or type /callout, to place the selected text in a box with rounded corners and a soft blue background that you can recolour. Callouts work inside list items and keep their look when a note is saved as Markdown. Code blocks in a note now show as a grey box in monospaced text, in both the editor and the note viewer.

2026.10.02 Databases > Open Rows, Selected References And Clickable Links

The control that opens a record now sits in its own leading column and appears when you hover over a row or reach it with the keyboard. Reference pickers mark the options already chosen with a checkmark, including when you reopen a value. Web links in URL cells open in a new tab, and clicking one no longer starts editing the cell.

2026.10.02 Files > Steadier File Downloads

Hardened file downloads: a brief network drop while a file is being fetched is now retried once automatically, and when a download is declined you now see the reason given instead of a general failure message.

2026.10.02 Notes > Edit Is One Click Away

Edit is now the first item in a note's right-click menu on the Notes list, card view and the Files page. In card view, hovering a note you can edit shows an Edit link that opens it straight in the editor. On a note you can only read, choosing Edit explains that you have read-only access.

2026.10.02 Desktop > Steadier App Start-Up

Hardened start-up in the desktop app. The update check now runs in the background instead of holding up launch, signing back in to your saved session has a time limit, and start-up as a whole always finishes on either your workspace or the sign-in screen, so the loading screen never waits indefinitely.

2026.10.02 Sharing > Change A New Collaborator's Permission Straight Away

When you share a project with someone from another organisation, the Shares tab now shows them by username as soon as the share is created, so you can change their permission straight away. If you shared by email address, the permission menu stays unavailable, with a note explaining why, until you next open the Shares tab.

2026.10.02 Notes > Deleting A Page That Holds Files

Deleting a page that still holds files stopped with a technical message. The app now tells you in your language how many files the page holds and that they need to be moved or deleted first, and the page stays in place.

2026.10.02 Chat > Saved Replies Shown After A Long Absence

When a chat checked back for an agent's answer after its tab had been away for more than an hour, it could show a failure note even though the answer had been saved. It now looks up that turn's saved reply and shows it in place, as it already does on a live connection, and a reply that is already on screen is never shown a second time.

2026.10.01 Chat > A Stopped Reply Keeps What It Had Written

When an agent's reply is stopped part-way, because you stop it or it runs too long, what it had already written is now saved in the conversation with a short note that it was stopped, instead of being lost. A stop before the agent has written anything saves nothing. The longest a single reply may run before it is cut off is now two hours rather than thirty minutes.

2026.10.01 Agents > Agents Stop When They Repeat The Same Call

An agent that keeps making the same tool call with the same result is now stopped after the third identical call instead of looping, and if it still repeats, the reply ends with a plain note saying it was stopped. A turn that has run for about 25 minutes is also nudged to wrap up.

2026.10.01 Chat > Replies Kept When A Device Sleeps Mid-Answer

When a phone or browser tab slept long enough for a reply's 30-minute wait to run out, the chat could show a failure note even though the agent had finished its answer. The app now checks for that turn's saved reply first and shows it in place; the note appears only when no reply was saved.

2026.10.01 Agents > Temporary Model Errors Are Retried

When the AI model hits a temporary problem on its side before any of the answer has appeared, the agent now retries it automatically, the same way it retries other passing failures. If the problem persists, you see a plain-words message asking you to try again in a moment instead of technical error text.

2026.10.01 Agents > Saving A File Outside A Project

Agents saving a file with an empty project value now save it to your Files instead of refusing the save. In a project chat, the file still goes into that project.

2026.10.01 API > Agent Jobs That Cannot Start Report Failed

A long agent job that stops before the agent can run, for example because the agent can no longer be loaded, now ends with state failed and a failure reason instead of finished with no reply. Billing refusals carry their own reason, and callers that do not stream get the matching status code, such as 413 for a message that is too large.

2026.10.01 Files > Steadier Uploads On Slow Connections

Hardened file uploads on slow connections: smaller files now get the same five-minute transfer window as the rest of the upload path, so a slow link can finish sending a file in one go.

2026.10.01 Chat > Branch A Conversation From Any Message

A message's actions menu now has Branch conversation. It starts a new conversation that carries the history up to that message, adds it to your conversation list and opens it, so you can take the discussion in a different direction without changing the original. If the message is still being saved, the app asks you to try again in a moment.

2026.10.01 Projects > Deleting A Project That Holds Other People's Items

Deleting a project that contains items other people own stopped with a technical message. The app now explains in plain words that those items belong to someone else and offers to move them back to the top level of each owner's Files, letting the owners know by email, before it deletes the project. Cancel leaves everything as it was. This works from the project's actions menu, from bulk delete and from right-click Delete.

2026.10.01 Files > Ask About iPhone Photos In A File Chat

Photos in the HEIC and HEIF formats that iPhones use by default can now be opened in a file chat, and the agent sees them just as it sees a JPEG. Your stored original is left unchanged, and a damaged photo gets a plain-words note rather than an error.

2026.10.01 Files > A File Someone Else Deletes Leaves Your List

When another project member deleted a file, it stayed in your file list. Opening it now tells you, in your language, that the file was deleted, and removes it from your list.

2026.10.01 Notes > Write With AI Asks You To Choose An Agent First

If no chat agent had been chosen yet in the session, Write with AI could send a request that failed with No agent available. Send now stays disabled until an agent is chosen, with a short hint in your language asking you to pick one first.

2026.10.01 Notes > Agent Edits Read On One Tidy Line

In a note's Agent Work section, each entry now starts with Agent edit: in bold, followed by a smart date such as Today, and stays on a single line that ends in an ellipsis when it runs long.

2026.10.01 Chat > Send A Screenshot On Its Own In A Project Chat

Pasting a screenshot with no text into some project chats could fail to send. Image-only messages now send in every chat.

2026.10.01 Chat > Opening A Conversation While The App Starts Up

Opening a conversation while the app was still starting could show an error loading the conversation. The app now fetches the conversation's agent and opens the chat.

2026.10.01 Chat > File-No-Longer-Available Message From The Sidebar

Opening a chat-with-file conversation from the sidebar after its file was removed now shows the same message as the chat itself, that the file is no longer available, and keeps you on the conversation you were in.

2026.10.01 Databases > Delete A Column From Its Menu

A column's menu in a database table now has a Delete column option. A short confirmation names the column first, and once you confirm, the column leaves the grid and the table's schema. It appears only on databases you can edit.

2026.10.01 Databases > File Columns With An AI Summary

Tables can now have a File column. Editing a File cell opens a dialog with two side-by-side choices: pick files you already have from your library, or upload a new one. The cell shows each file as a chip you can click to open its details. Switch on Show AI summary for the column and a read-only summary column appears right next to it. A row that links an existing file also shows that file's name as a link to its details.

2026.10.01 Databases > Pick Dates And Times From A Calendar

Date-and-time cells now open a real date and time picker instead of a free-text box, so a value can only be a real date and time. Values display in your local format.

2026.10.01 Databases > Edit A Whole Row In One Form

Every row now has an Open button that shows all of its properties in one form, editable and saved together. The toolbar's + Row button opens the same form for a new row. The Add New Row link now sits inside the grid, directly under the last row, and adds a blank row straight away with no form, ready for typing.

2026.10.01 Databases > Table Tags, Clearer Labels And A Tidier Grid

A table's tags now show as chips beside its tab and in the sidebar. The tab strip is labelled Tables and the toolbar Actions, and each sidebar entry carries a Table or View chip. The grid has a single border and runs to the edges of the screen. The table menu shows every action without scrolling, with Delete table set apart. When you choose a reference column's target table, the selected one is highlighted with a check.

2026.10.01 API > Database File Columns And Stricter Date-Time Cells

Database tables accept a new file column type through POST /v3/databases/schema, configured with multiple and show_summary. A file cell holds a list of file uids and reads back as file chips carrying uid, access, and, when you can read the file, its display_name and AI summary. File columns support only the is_empty filter and cannot be sorted or grouped. Date-time cells must now be ISO-8601 date-times; any other value is refused. Database tables can also be tagged through POST /v2/srm/tag.

2026.10.01 API > Redeem A Public Agent Link

A new endpoint, POST /v2/sharing/redeem-public-link, lets a signed-in user redeem an agent's public link token. It returns the agent's uid and whether the user already had access, and adds the agent to that user's agent list. POST /v2/sharing/create-share now accepts resource_type "agent" with scope "public_link", and POST /admin/register-new-user accepts an optional agent_share_token so a new account can be created straight from an agent link.

2026.10.01 Accounts > Sign Up Straight Into A Shared Agent

Opening a shared-agent link without an account now leads to a short signup page, Access Shared Sapience Agent, that asks only for first name, last name and email. After Chat with Shared Agent, the new user lands in a focused chat with just that agent: no sidebar and no other agents. Opening the link while already signed in adds the agent to your agent list and takes you straight into the chat.

2026.10.01 Sharing > Publish An Agent As A Public Link

The owner of a personal agent, or an org admin for an org agent, can now publish that agent from the chat header. The Published row shows whether the agent is published and offers Copy URL. The link lets anyone sign up and chat with that one agent. Links can carry an expiry date and a maximum number of uses. Unpublishing ends access for everyone who joined through the link.

2026.09.30 Outbound > Lists Page Split Into Three Tabs With A Readable ICP Window

The Lists page in the Outbound Ops Console now splits synced LinkedIn lists, live provider connections and saved lists into three tabs, shows one view at a time, and keeps the chosen tab in the address so a reload or shared link opens it again. Synced networks open from a compact roster with their key numbers as tiles, live connections appear once you pick an account, and each saved list's ICP opens in a window that shows its criteria as plain labels, chips and ranges in place of raw data.

2026.09.30 Databases > Saved View Filters Read In Plain Words

The Databases sidebar showed a saved view's filter as raw code. It now summarises the filter in words, using the table's own column names, puts nested groups in brackets, and appears in every language the app supports.

2026.09.30 Projects > Save To Project Works From A Project Chat

Choosing Save to project on an agent's reply inside a project chat could fail with an error. It now saves the conversation you are looking at as a note in the project you pick, and offers to open that project.

2026.09.30 Notes > Hardened Duplicating A Note

Hardened duplicating a note, including when a new version of the app has been released while you have it open. The app now retries, reloading itself if it needs to, in place of showing a failure.

2026.09.30 Files > Slide Previews Keep The Rest Of A Grouped Shape

Slide previews now skip a single shape they cannot identify, such as some SmartArt, and still show the rest of its group and the rest of the slide.

2026.09.30 Databases > Explore A Read-Only Demo Database

A read-only Demo Database now appears in everyone's Databases list, marked with a Global scope chip. It is a Northwind-style sample with seven linked tables (categories, suppliers, products, customers, employees, orders and order lines), a saved view on every table, reference columns that link rows across tables, and lookups that pull values through those links. Browse, sort and filter it to see what Sapience Databases can do before you build your own. Only its owner can change it.

2026.09.30 API > New Webhook Event When A Database Row Is Added

Outbound webhooks can now subscribe to domain_databases_row_added, which fires once for every row actually added to a database table. For a new row the event carries database_uid, table_uid, row_uid, position, created_at_utc, added_by_user_id, row_kind "created" and the row's cell values in data, with reference cells given as links carrying the referenced row's uid and type. When the row is an existing Sapience object linked into the table, row_kind is "linked", data is null and linked_resource says it is a link to an existing resource with its uid and type. Re-linking an object that is already in the table sends nothing. The same event is also pushed over the app's realtime connection to the database owner and to the person who added the row.

2026.09.30 Databases > Add Rows And Columns Right From The Grid

In a database table you can now add a row without going back to the toolbar: an Add New Row link sits directly under the rows and opens the same new-row form as the toolbar button. A blue + button sits immediately to the right of the last column header, with a tooltip, and opens the same Add Column form as the toolbar. Both appear only on databases you can edit. The toolbar and table-tab actions are now blue links in normal case rather than all-caps buttons.

2026.09.30 Databases > Tidier Database List With Scope Chips

The database list on the left of the Databases screen now opens with an All Databases heading. Each database is a collapsible item, and only the database you are working in is expanded, so its tables, views and actions are the only ones on show. Every database card carries a chip naming its scope: User, Team, Org or Global. New and New View appear as blue links in normal case, while Rename, Tag, Move and Copy stay as buttons. The new labels are translated into every language the app supports.

2026.09.30 Outbound > Enrich People Straight From Prospect Results

The Prospect People results in the Outbound Ops Console are now a grid showing every field the search returns, including the organization name, location and last-refreshed date. Wherever the search knows a value exists but has not revealed it, the cell shows a blue underlined enrich link. Clicking it opens a dialog where you choose Fast, which uses a single source with a small spending cap, or Full, which runs the regular enrichment flow. The enrichment is queued as a background job, the dialog shows progress and can be closed while it runs, and a toast appears when it finishes with the revealed values filled into the row. Whatever is found is saved to the outbound graph, so later enrichment runs skip anything that is already fresh, and enriching the same person again costs nothing.

2026.09.30 Agents > Files Added To An Agent Show Their Status

Saving an agent with new files now closes the dialog within a few seconds and adds the files in the background. On the agent's Knowledge Base tab each file shows Queued, Adding…, Ready or Failed, and the status updates live as each file finishes. A file that could not be added shows the reason and a Retry button. Ready means the file is attached to the agent; its search index can take a few more seconds to complete. The Regenerate knowledge library dialog on a project can also be closed while the rebuild carries on.

2026.09.30 Agents > Hardened Adding Large Batches Of Files To An Agent

Hardened adding many files to an agent at once. The app now sends files in groups of ten, sends each group once, and keeps every file in a single knowledge base for the agent however many groups are in flight. Spreadsheets added in parallel are consistently kept for data analysis, and tab-separated (.tsv) files are now treated as spreadsheets, the same as csv, xls and xlsx.

2026.09.30 API > Add Files To An Agent Without Waiting For Indexing

POST /api/v2/agents/{uid}/add-file and /add-file-bulk accept wait_for_indexing=false, which returns as soon as the files are accepted: add-file answers 202 with the message queued, and add-file-bulk returns each file's status (queued with a job_uid, or ready if the file is already attached) and a queued count. GET /api/v2/agents/{uid}/list-files accepts include_attach_status=true, which adds attach_status (queued, adding, ready or failed) and attach_error to each file. A realtime event, domain_agent_file_attach_settled, tells the user who added a file when it is ready or has failed. Without the new parameters both routes behave as before.

2026.09.30 Notes > See An Agent's Work On A Note

When an agent has worked on a note, the note's floating panel in the viewer and the editor now has a third section, Agent Work, below Headings and Comments. It shows the agent's notes on what it changed, newest first, as formatted text, with a line naming the agent's status and which agent last edited the note. Long logs scroll inside the section. Notes no agent has touched show no Agent Work section.

2026.09.30 Chat > Clear Agent/Skill Combo Returns Home To A Fresh Start

Picking an agent, a skill or an inspiration card on Home hides the Inspiration Gallery. A blue Clear Agent/Skill Combo button now appears when that happens, and clicking it resets Home exactly as if you had just opened it: the gallery returns and the chosen agent, skill, connected app, Advanced options, typed text and staged attachments are all cleared.

2026.09.30 Chat > Wide Tables Stay Inside The Chat Bubble

On desktop, a wide table in an agent's reply could extend past the right edge of its chat bubble. The table now stays inside the bubble and scrolls sideways within it, as it already did on mobile.

2026.09.30 Connected Apps > Hardened Connected App Setup For Microsoft Email

Hardened how Connected Apps finds each app's configuration when you connect it, so every configured app is found however many are set up, including Microsoft Email (Outlook).

2026.09.30 Files > Videos Without A Sound Track Report No Speech

Transcribing a video that has no sound track now reports that the video contains no extractable text, in place of a transcription error, and no summary is written from an empty transcript.

2026.09.30 Desktop > Sapience Toolbar Adds Frequent Items, Favorites And One New Menu

The Sapience Toolbar has two new rows: Frequent Items, which opens the same search as Search Sapience, and Favorites, which brings the main window forward with your Favorites dialog open. The three separate New rows are now one blue New button that opens the same seven choices as the app's New menu — Conversation, Note, Project, Task, Agent, Meeting and Team Chat — and clicking the collapsed S opens the main window directly. The Favorites dialog itself now shows one row per line, so about twice as many favorites fit on screen.

2026.09.30 Skills > Skill Runs Stay Connected For Longer Tasks

Messages sent with a skill selected, whether picked from the menu or typed as a /skill name, now use the realtime connection whenever it is available and fall back to long-polling otherwise. Long-running skill work is no longer tied to a single streaming request, so extended skill runs complete reliably.

2026.09.30 Accounts > Account Status Check Shows A Clear Temporary Message

When Sapience cannot confirm your account status for a moment, it now says the service is temporarily unavailable and to try again in a moment, instead of saying that payment is required. A real payment-required refusal keeps its payment message.

2026.09.30 Desktop > Installed Desktop App Runs The Newest Web App

After you install a desktop update, the app now runs the newer of the web app it shipped with and the copy it downloaded in the background, instead of keeping an older downloaded copy under the same version. The Sapience Toolbar New items now open the new item in the installed app instead of only bringing the window forward.

2026.09.30 Skills > RADAR Assessor Skill Listed Under EFQM

The RADAR Assessor skill now appears under the EFQM group in the Skills menu instead of under Assessment. It is also shown without having to be installed first, so it is no longer missing from the EFQM submenu.

2026.09.30 Outbound > Sender Picker Lists Seats For Console Campaigns

In the Outbound console, the sender picker for a campaign created there now lists the available LinkedIn seats, where it used to say that none were available. Campaigns that already carry a channel plan behave as before, and picking a seat never starts sending by itself.

2026.09.29 Chat > Enter Sends Messages That Contain An Email Address

Pressing Enter on a message that contained an email address could do nothing, because the @ opened the agent list with no matching agent. When no agent matches, the list now closes and Enter sends the message. Shift+Enter still adds a new line, and Enter still picks the highlighted agent when one matches.

2026.09.29 Projects > Task Details Show When A Task Is Not Yours To Edit

Project members who did not create a task could change its fields in the task details, and Save then refused the change. The task's own fields now open read-only for anyone who can't edit that task, with a line explaining why. The project owner and admins can still edit every task, and linked files can still be changed.

2026.09.29 Agents > More Reliable Presentation And Spreadsheet Creation By Agents

Hardened the tools agents use to create presentations and spreadsheets. Only a file produced by the current request is saved as its result, and a request that produces no file is reported as a failure rather than a success.

2026.09.29 Projects > Files Agents Save Into A Project Reach The Project Agent

Slide decks, images, generated files and quick saves that an agent writes into a project, including into a folder inside it, are now added to the project agent's knowledge, the same as files you upload, so the project agent can find and use them.

2026.09.29 Agents > More Reliable Spreadsheet Analysis And Editing By Agents

Hardened the tool agents use to analyse and edit Excel workbooks. The agent now receives the model's written analysis directly, only a workbook produced by the current request is saved as its result, and the output file is recognised however the model formats its name, including names that contain spaces.

2026.09.29 Accounts > Sign Up With Any Email Address

Sign-up now accepts any valid email address, including personal addresses from providers such as Gmail, Yahoo, Hotmail, Outlook and AOL. Previously those addresses were refused with a request to use a work email. Addresses that are not correctly formed are still rejected.

2026.09.29 Chat > Copy Falls Back To Plain Text

Copying a message could fail in browsers that refuse a formatted copy. Copy now falls back to the message's plain text, so the text still reaches your clipboard, and a failure is reported only when neither copy succeeds.

2026.09.29 Chat > Conversation Menu Waits Until A New Conversation Is Saved

Choosing an action from a new conversation's menu before the conversation had finished saving, such as Delete, could show a misleading failure message. Until the conversation is saved, each action in that menu now shows a short message saying it is still being saved, and nothing is changed.

2026.09.29 Chat > Clear Message When A Model Can't Read Images

Sending an image to a model that can't read images now returns a plain message naming the model and suggesting a few models that can, or sending the message without the image, in place of a technical error.

2026.09.29 Projects > Adding A File Whose Name Is Already In The Project Names It

Adding a file to a project that already holds a different file with the same name now shows a message naming that file and suggesting you rename one of them or open the existing one, in place of a count of failed files. File names containing a dollar sign now appear exactly as written in these messages.

2026.09.29 Projects > Uploads Into A Project Finish Sooner

Uploading a file smaller than 10 MB into the top level of a project no longer waits on a separate step that repeated work the upload already does, so each upload finishes sooner and frees its place in the upload queue earlier.

2026.09.29 Projects > Deleting A Project Switches Off Its Scheduled Jobs

Deleting a project now also switches off the scheduled jobs set up for it, so they no longer run once the project is gone. The jobs are disabled rather than deleted. A scheduled job that fires after its project has been removed now switches itself off.

2026.09.29 Notes > Agent Edits Record Their Changes Outside Your Note

When an agent edits a note, its record of what it changed is now kept in the note's notes_from_agent_md field, newest entry first, instead of being added to the top of the note's text. Saving the note in the app keeps those entries.

2026.09.29 Chat > Home Sends The Full Advanced Option You Pick

Picking an option in Advanced options on Home now sends its reasoning effort and temperature along with its model, the same as picking it inside a conversation. A Thinking option chosen on Home now runs at the effort it names.

2026.09.29 Connected Apps > Starter Chats Run On Gemini 3.8 Flash

Starter questions and Start chat on the Connected Apps pages and the Home Connected Apps module now run on Gemini 3.8 Flash, Google's newest Flash model.

2026.09.29 Models > Anthropic Claude Sonnet 5.5

Anthropic's Claude Sonnet 5.5 (model code anthropic/claude-sonnet-5-5) is selectable in the chat model picker, the create-agent wizard, Advanced options, and the AI Store model browser. It carries a one-million-token context window, up to 128,000 tokens of output, and reads images. Pricing is $2 per million input tokens and $10 per million output, with cached input at $0.20 per million. Reasoning is always on and cannot be turned off at any effort level. The chat model picker now lists Sonnet 5.5 as its Claude Sonnet option; Sonnet 5 is still available in the create-agent wizard, Advanced options, and the AI Store model browser. The Sapience-built agents that ran on Claude Sonnet now run on Sonnet 5.5.

2026.09.29 Chat > Home Uses Any Model You Choose In Advanced Options

A model chosen in Advanced options on Home is now applied to the message you send, including models that are not on the chat model picker, such as Claude Sonnet 5 or Claude Opus 5. Opening a chat link that names one of those models also keeps it.

2026.09.29 Desktop > Restart To Update Installs The Newest Version

Restart to Update now checks for the newest release at the moment you click, so the desktop app moves straight to the latest version even if several releases shipped since the update was first downloaded.

2026.09.29 Desktop > More Reliable Background App Updates

Hardened the desktop app's background update installation, improving reliability when applying larger updates.

2026.09.29 Agents > An Agent Can Find Its Tools On Demand

An agent can now be set to look up its tools as it needs them, instead of receiving its whole toolset at the start of every conversation. Set tool_search_first to true in the agent's mba_agent_config, saved through POST /v2/agents/update/{uid}, and the agent starts each new conversation with a smaller core set of tools and a short summary of what it can reach. It then calls find_tool to search the tools, agents and skills it is permitted to use, and uses what it finds on its next step. Searching never widens what an agent is allowed to do. When the agent hands work to another agent, that agent follows its own setting. The setting is off by default, so existing agents keep receiving their full toolset.

2026.09.29 Chat > Opening A Conversation Link After Your Sign-In Expires No Longer Shows A False Error

Opening a link to a conversation after your sign-in had expired could briefly show "Unable to load the selected conversation." on the way to the sign-in screen, even though nothing was wrong with the conversation. That message no longer appears, and you simply land on the sign-in screen. Switching quickly from one conversation to another also no longer lets a late error from the first appear over the second or return you to an empty chat.

2026.09.29 Desktop > Sapience Toolbar Labels Voice Agent, Drops Underlines

In the Sapience Toolbar fly-out the rows are no longer underlined, and the Voice row is now labelled Voice Agent.

2026.09.29 Agents > EFQM Management Report Print Layout Refresh

The EFQM Management Report now prints with a refreshed layout: the Rubik font, a gradient cover and back cover, a cover that shows the report title and the organization name, full-page pillar covers, each criterion starting on its own printed page, and the stakeholder weighting workbook shown as a cleaned table inside Criterion 3. Duplicate overview headings are removed, and the report footer shows the assembler version so old and new files can be told apart.

2026.09.29 Organizations > Child Organizations See Files From Shared Parent Projects

Improved how files appear in organization projects: members of a child organization now see the files inside a parent organization's project that is shared down to child organizations, and a member of the same organization sees a colleague's file inside an organization project, in the project's file tree and listings.

2026.09.29 Chat > Home Inspiration Gallery Of Agent And Skill Ideas

Home now shows an Inspiration Gallery between the greeting and the agent picker: a small rotating gallery of four ideas, each naming a real agent and, where one exists, a skill. Clicking a card selects that agent, installing it from the AI Store if you do not have it yet, and places the skill and a starter prompt in the message box without sending it. It pauses on hover, can be stepped with arrows, and stays still if your device asks for reduced motion.

2026.09.28 Sharing > A Refused Share Tells You Why

When a share is refused, for example because only a file's owner can share it, the message on the file card and on a project's Shares tab now includes the reason, in place of a generic "Failed to create share".

2026.09.28 Files > Pasting A Folder Where That Name Exists Gives A Plain Reason

Pasting or copying a folder into a place where you already have a folder with the same name is still refused, but the message now says plainly that a folder with that name already exists there, in place of a technical error.

2026.09.28 Agents > A Hand-Off To Another Agent Records What That Agent Ran

When one agent hands a task to another agent, the parent conversation's record of that hand-off now carries the other agent's run identifier and a list of the tools it called, each with whether it succeeded. This is recorded even when the hand-off fails part way. What the parent agent receives as the result is unchanged.

2026.09.28 Skills > An Armed Skill Can Bring The Tools It Needs

Arming a skill in a chat now also gives the agent, for that turn only, the tools the skill says it needs, provided Sapience has marked those tools as ones a skill may grant. Previously a skill could describe a tool that the agent in your chat did not have. Tools that are not marked grantable are never added, the agent's saved configuration does not change, and the rest of its tools stay exactly as they were. The EFQM Management Report skill is the first to use this.

2026.09.28 API > Finalize An EFQM Management Report, And Generate Returns Three Files

POST /v2/utilities/tools/execute with tool_name "generate_efqm_report" now writes three files to the input folder, the report as JSON, an applicant HTML copy and an assessor HTML copy, and returns the uid of each. A new tool, "finalize_efqm_report", takes an efqm_uid and marks that report final and read-only; generating again on a folder whose report is final is refused with a clear message. The report record now points at the JSON file rather than holding the report inline, and records written by earlier versions remain readable.

2026.09.28 Agents > EFQM Management Report Produces An Applicant Copy And An Assessor Copy

Every time the EFQM Management Report is assembled it now saves two documents in the input folder, alongside a JSON file holding the full report. The applicant copy contains what the applicant submitted and the model's own headings, with no AI-authored text at all. The assessor copy adds the AI flags, cross-references and related Sub-Criterion notes, each clearly labelled as AI-authored. Both carry a footer naming the assembler version and the run that produced them, so it is always clear which run a document came from. The report's record checks the JSON file's contents every time it is read, so a file changed after the fact is reported rather than trusted, and assembling the same folder again creates a new version of the record without picking up its own earlier outputs as input.

2026.09.28 Agents > EFQM Management Report Handles More Of What Applicants Submit

Hardened the EFQM Management Report assembler to accept submissions as applicants actually prepare them. Every indicator a Results workbook contains now appears in the report, however many indicator sheets it holds. Indicators graded with letters or words instead of figures are shown exactly as entered, in a table without a chart. Notes that carry Sapience's own edit history above the applicant's content assemble normally, with that history left out of the report and recorded as excluded. Spreadsheets supplied as additional files appear as tables with their cell text as entered, files attached to a note are listed by name under that note's Sub-Criterion as supporting documents, and the cover and the Direction, Execution and Results dividers use the EFQM 2025 model colours. A run never modifies the applicant's files.

2026.09.28 Projects > Refresh From Cloud Replaces A Project's Local Data With The Cloud's

Hardened project refresh so it is a true reset to what is in the cloud. Refresh from Cloud in a project's Actions menu or header, and the refresh button on the project's Files & Notes and Files tabs, now always make their own fresh read of the whole project and replace what you see with exactly what the cloud returned, so files, folders and notes removed elsewhere drop out of the view. Anything you create while a refresh is still running is kept, and a refresh that cannot complete leaves your current view as it was rather than removing anything. The Tasks tab refresh continues to update tasks and goals only.

2026.09.28 Projects > Opening A Project Always Loads Its Latest Contents

Opening a project shows what Sapience already has straight away, then always fetches the project's full, current set of files, folders, notes, tasks and goals from the cloud. This applies when you open a project, switch to another one, or follow a link to it, so changes made by teammates and agents since your last visit appear without a manual refresh.

2026.09.28 Search > Frequent Tab Counts Only What You Actually Open

The Frequent tab in search now counts an item only when you open it in a window you can see. Reloading the page, a window restoring itself in the background, or the same account being open in several tabs, browsers and devices no longer adds to an item's count, and a note opened in a preview now counts the same as opening its page. Something you open into a background tab counts when you switch to that tab. Your Frequent list starts fresh once with this update, then builds from how you actually use Sapience.

2026.09.28 Desktop > Sapience Toolbar No Longer Appears In The Linux Taskbar

On Linux desktops the Sapience Toolbar showed up as its own entry in the dock or taskbar, next to the main Sapience window. The toolbar is now created as a toolbar window, so the desktop no longer lists it there and it stays on top as before. Windows and macOS are unchanged.

2026.09.28 Chat > Deleting The Open Conversation Reports Success

Deleting the conversation you had open could say "Failed to delete conversation" even though it had been deleted, and trying again then said the conversation could not be loaded. The delete now reports success, the app stops pointing at the deleted conversation, and no empty replacement conversation is created.

2026.09.28 Chat > Clear Message When A File Chat's File Is No Longer Available

Opening a chat-with-file conversation whose file has been removed, or is no longer shared with you, now explains that the file is no longer available and the file chat can't be opened, in place of a technical message. Other temporary problems opening a file chat ask you to try again in a moment.

2026.09.28 Agents > Adding Many Files To An Agent Runs Several At Once

Adding a batch of files to an agent's knowledge now processes up to six files at a time instead of one after another. Each file still gets its own result, in the order you sent them, and one file failing never stops the rest.

2026.09.28 Projects > Project Menu Shows Every Action And Explains Permissions

The right-click menu on a project card and on the project page now always shows Delete and the organization publishing actions that apply to the project. If you don't have permission for one of them, choosing it opens a short explanation of who can do it, rather than the item being missing. Only a project's owner can delete it.

2026.09.28 Files > Resumable Uploads With A Name That Already Exists Are Refused Up Front

Resumable uploads now get the same same-name check as other uploads, both when they start and again when they finish. An upload into a folder or page that already holds a file with that name, ignoring case, now stops with a message saying so, where previously it could fail only at the end and leave an extra copy of the file in your root folder.

2026.09.28 Agents > Coding Sandbox Runs Report Their Time Limit And Fit The Agent's Turn

When a coding sandbox run reaches its time limit, the agent is now told plainly that the limit was reached and that the task should be split into smaller runs, so it can explain this and carry on. A sandbox run started late in a long agent turn is given only the time that remains in the turn, and a run is not started at all when too little time is left.

2026.09.28 Files > Folder Tree Does Less Work For Accounts With Many Shared Items

Opening a folder or project in the folder tree now checks only the sharing records that can affect that folder, instead of every item that has ever been shared with you. The top-level tree now prepares sharing details only for the items it is showing. What you can and cannot see is unchanged.

2026.09.28 Search > Clicking The Sidebar Search Box Opens Search Directly

The search box in the sidebar used to show a Sapience/Web switch next to it. That switch is gone: clicking into the box now opens the full search window right away, starting on Sapience search. Web search is still one click away from inside the search window itself.

2026.09.28 Desktop > Sapience Toolbar Gains Icons, Open Sapience And Voice

Every action in the Sapience Toolbar's fly-out now shows its icon beside the label: chat, the Sapience note S, dictate, project, search and a microphone for voice. The icons come from the Sapience app itself, so they always match the ones you see in the main window. Two new rows lead the list: Open Sapience brings the main window to the front, and Voice starts or stops a voice conversation with your agent, exactly like its keyboard shortcut. If the main window is still starting up, the rows show as text and the icons appear the next time you open the fly-out.

2026.09.28 Files > One Note Icon Across Files

Sapience notes now use one icon, a white square with a black S, everywhere in Files. Note rows in the Files list and grid, the Show Notes toggle, the file details card and the file picker previously showed a black square instead, so a note now looks the same in every one of those views.

2026.09.28 Databases > Tables, Named Rows And Linked Columns In The Databases Screen

The Databases screen in the web app (a preview, opened at /#/databases) now works with databases that hold many tables. Each database shows its tables as tabs you can add, rename, reorder, tag, duplicate and delete, and every row has a Name as its first column. Columns can be any of 17 types, including currency, percent, rating, duration, email, URL, phone and date-time, each with its own editor and formatting. A reference column links rows to a table in any database you can read, shown as chips you can click to jump to the linked row, and a lookup column pulls a field from the linked row and stays current when that row changes. Expanding a row lists every row that points at it, and deleting a row that others still reference tells you how many references remain. Pressing Enter submits every dialog on the screen.

2026.09.28 Databases > Totals, Grouping, Pivot And Export In Database Grids

Database grids now load the whole table (up to 5,000 rows) so totals describe the table rather than the page you are looking at. Group rows by a saved view or by dragging a column into the grouping bar to get subtotals and a grand total, switch to pivot, filter by a column's values, select a range of cells to see its sum in the status bar and fill it down, and export to Excel or CSV. Saved views on each table keep their filters, including filtering by a linked row or by rows with no link, their sort, grouping and a chosen set of visible columns.

2026.09.28 Chat > Sending A Message Scrolls To The Bottom Of The Transcript

When you send a message after scrolling back through a conversation, the transcript now jumps to the bottom so you can see what you just sent and the beginning of the reply. That includes sending while an answer is still streaming, where your message is queued and then delivered. Messages that the app sends for you later, such as a queued message going out when the turn finishes, deliberately leave your position alone so the view never moves at a moment you did not choose.

2026.09.28 Files > Faster File Lists For Accounts With Many Shared Items

The Files list now reads only the sharing records that apply to the items on the page it is building, instead of the whole set each time. Accounts with large numbers of shared folders and projects see the biggest improvement, and what you can and cannot see is unchanged. Hardened the same path so a very large page of files can no longer exceed the database's limit on values in a single query.

2026.09.28 Accounts > Logging Out While The App Is Starting Finishes Cleanly

Logging out while the app was still starting up, or while it was loading your conversation list in the background, could stop startup on an internal error. Startup now finishes cleanly in that case, and the background conversation load simply stops once you have signed out.

2026.09.28 Desktop > One-Line Install And Upgrade For The Linux Desktop App

Linux users can install or upgrade the desktop app with a single terminal command that downloads the latest version, adds a menu entry and a desktop shortcut, and needs no administrator rights; running it again upgrades. Download for Linux in the Desktop apps window now opens a chooser with two options: download the AppImage, or copy the one-line command, which has a copy button beside it. The Linux package is also smaller and starts in about half a second instead of over a second.

2026.09.28 Agents > Fresh Coding Workspaces Start Reliably

Hardened start-up of fresh coding workspaces for agents that build presentations, including the refresh of a workspace's stored copy, so new workspaces start normally.

2026.09.28 Files > File Summaries Are Saved Whatever Characters They Contain

Hardened saving of a file's AI summary, so a summary is stored even when the generated text contains unusual characters.

2026.09.27 API > Outbound Webhooks For Projects, Tasks And Scheduled Jobs

Webhooks can now subscribe to seven more events beyond a new user being created. For projects: project_status_changed (completed, archived or reactivated) and project_file_uploaded. For tasks: task_created, task_due_date_reached, task_status_changed and task_assignee_changed. For scheduled jobs: scheduled_job_finished fires when a job completes or fails and carries its status, timing and result summary; for an agent query job it also carries the agent's answer as output_text (up to 16,000 characters, with output_truncated set when it is cut) plus the conversation_uid and agent_uid. Register them with the same endpoints: GET /api/v2/webhooks/eligible-events lists every event and the fields each one carries, and POST /api/v2/webhooks/create accepts the new names. Each event is delivered to the webhooks of the project's, task's or job's owner and, when it belongs to an organisation, to that organisation's webhooks.

2026.09.27 Notes > Resize Embedded Content In Notes

Embedded content in a note, such as a Google Sheet, can now be resized. Select the embed and drag its bottom edge to set the height you want. The height is saved with the note and used wherever the note is shown, including read-only views and shared pages.

2026.09.27 Desktop > Sapience Toolbar Opens On Hover And Hides From Its Own Menu

The Sapience Toolbar's fly-out now opens when you hover over the S as well as when you click it, and it uses a solid panel that follows your light or dark theme under a Sapience Toolbar heading. You can hide the toolbar by right-clicking the S or from a Hide Sapience Toolbar row at the bottom of the fly-out, which uses the same setting as the tray menu. The New Chat tooltip shows the Quick Chat shortcut you have configured.

2026.09.27 Sharing > Deleted Files And Conversations Clear From Shared With Me

When a file or conversation that was shared with you is deleted, it now drops out of your Shared with me list straight away, matching how deleted projects already behave.

2026.09.27 Menu > Favorites Load Reliably When You Sign In

Hardened how your favorites load when you sign in, including on a new device and when you sign in again in the same tab. Loading retries automatically on a temporary failure, the sidebar shows that favorites are loading, and if they still cannot be fetched it offers a Retry that fetches them again without reloading the page. A change you make while favorites are still loading is always kept.

2026.09.27 API > Per-Request Reasoning Effort Without A Model Override

API callers can now set how hard the agent thinks on a single request without overriding its model: send a reasoning effort from Low to Max with no model attached and the agent's own configured model runs that turn at the requested effort. Leaving the effort unset keeps the agent's own setting, an empty override is refused, and sampling temperature is left untouched unless explicitly provided.

2026.09.27 Desktop > Sapience Desktop App Now Available For Linux

Sapience now ships a native desktop app for Linux, alongside Windows and Mac — download it from Menu ▸ Apps. It brings the full desktop experience: global keyboard shortcuts and system-tray access, automatic two-way file sync, seamless login, auto-updates, and multi-window support.

2026.09.27 Desktop > Quick Chat Stays Open In The Background And Starts A Fresh Chat Each Time

Closing Quick Chat now tucks it away in the background instead of shutting it, so the next Ctrl+Shift+S or toolbar New Chat brings back the same window instantly and starts a new conversation without reloading. Hardened how Quick Chat picks up your existing sign-in, so it reuses the session the main window already has on every platform, including Linux, and it follows you when you log out or sign in as someone else.

2026.09.27 Desktop > Log Out Completes Promptly In The Desktop App

Hardened desktop Log out so it finishes promptly and returns you to the sign-in screen, including when you log out from inside Quick Chat, and the app does not sign you back in automatically the next time it starts.

2026.09.27 Chat > Reasoning Effort Control In Advanced Chat Settings

Advanced Chat Settings now has a Reasoning selector, so you can set how hard the model thinks on a single request: None, Low, Medium, High, Extra High, or Max. Your choice overrides the selected model's built-in effort for that request, and leaving it on the default keeps the model's own setting. The selector is available whenever you override the model for a request.

2026.09.27 Agents > Specialists Answer Quick, Balanced, Or In-Depth When Asked By Another Agent

When one agent asks another agent to handle part of a task, it can now say how thorough the answer should be: a quick reply for a simple question, the usual balanced detail, or a long in-depth response for serious work. Saying nothing keeps the previous behaviour, so the specialist answers with balanced detail as before.

2026.09.27 API > Databases Now Hold Multiple Tables

A Sapience database was a single table: its columns belonged to the database itself and a view was a subset of those columns. A database can now hold many tables, the way a workbook holds sheets. POST /api/v3/databases/schema, /rows and /query all take a table_uid naming the table you mean, columns belong to a table rather than to the database, and a view is a column subset of one table. Each row also carries its own name, so a row reads as a named object rather than an anonymous set of cells. Existing databases have been reset for this change.

2026.09.27 API > Link Rows Across Databases With Reference Columns

A column can now be given the type reference, which points a row at rows in another table, including a table in a different database. A reference cell reads back as a list of chips, each carrying the target's name together with the identifiers you need to follow it, so you can render a link without a second request. The name is resolved when you read: rename the target row and every chip pointing at it shows the new name immediately, with no write to the rows that refer to it. Filtering by a reference is supported, and deleting a row that other rows point at is refused with the number of references still pointing at it, so you can decide what to clear.

2026.09.27 Agents > Per-Agent Coding Workspace Settings Take Effect

An agent's own coding-workspace settings are now honoured when it runs code in its workspace, in place of the global defaults. Settings that are not valid are rejected.

2026.09.27 Desktop > Saved Sign-In Survives A Temporarily Unavailable Keyring

Hardened saved sign-in on the desktop apps. If the system's secure storage is only temporarily unavailable at start-up, for example not yet unlocked, you see the sign-in screen that time and your saved sign-in is restored at the next start.

2026.09.27 Projects > Failed Remove From Project Shows A Readable Message

When removing a file from a project fails, the message you see is now one readable sentence instead of the raw technical error text.

2026.09.27 Skills > Skills Load Freely But Run Only With Approval

Agents can now read a skill's instructions so they can make an informed offer, but they run its steps only after you ask for the skill, confirm it, or an approved plan names it. A skill that chains to another named skill needs the same approval.

2026.09.26 Outbound > Prospect List Builder In The Outbound Ops Console

The Outbound Ops Console now has a Prospect area for building prospect lists. Search people by job title (quoted titles match exactly, with an option to include similar titles), management level, location, company headquarters location, company size, industry, keywords and more, or search companies and jump from any company to the people who work there. A Lookalike tab takes up to ten LinkedIn profiles and finds people like them, and its filters can be copied into a people search. Your search reads back as a plain sentence with a live match count, such as 4,349 people who are chief financial officer or CFO, based in California, at companies with 101 to 2,000 employees; each part can be removed with one click, and the search lives in the page link so it can be shared. Lock the search to review it, then save it to a list and optionally queue enrichment for every member, which runs in the background while the page shows when the list is saved and links straight to it.

2026.09.27 Desktop > Install The Desktop Sidecar (Experimental)

Desktop Settings now has an Install sidecar (experimental) link. Clicking it downloads a small self-contained Bun program for your computer, checks it, starts it on a free local port and remembers its version, location and port. If the installed version is already the latest, the app starts the copy you have instead of downloading again. A Test Connection button then calls the sidecar's hello endpoint, starting the sidecar first if it is not running, and shows the result with error details if it cannot connect. The sidecar only listens on your own machine and closes when the app closes.

2026.09.27 Chat > New Chats Started From Links And Shortcuts Stay In The Conversation You Opened

Hardened how a new chat is started from Home, search, a project page, a link or desktop dictation, so a message always goes into the conversation you are looking at. Leaving the page, cancelling a prompt, or a page change that fails partway through now releases the pending new chat instead of keeping it for a later message, and desktop dictation no longer leaves an empty conversation behind.

2026.09.27 Files > Moving Items From Several Windows At Once Keeps Folder Counts Accurate

Hardened moving files and folders when several windows or people are moving items at the same time. The folder item counts that follow a move now retry automatically on a temporary conflict, and moving a whole folder tree completes cleanly or retries as one unit.

2026.09.27 Projects > Deleted Projects Clear From The Project List And Shared With Me

Hardened how deleted projects are cleared across the app. Deleting a shared project now removes it from Shared With Me for everyone it was shared with straight away, and when the project list loads Sapience confirms any shared project that may have been removed and clears it from the list and from the app's saved copy, while a project that is shared with you again later still appears.

2026.09.26 Agents > Agents Emailing You Without Naming A Recipient Send To You

When an agent is asked to email you and leaves the recipient blank, it now sends the email to you directly on the first try.

2026.09.26 API > Batch Move, Copy And Delete Report Every Item And Accept Up To 500 Items

POST /api/v2/srm/move, /copy and /delete now accept up to 500 uids in one request; a larger request is refused with a 400 and nothing changes. Results come back in the order you sent the uids, with a reason on each item that could not be processed while the rest go ahead. When every item in a request of two or more fails, the response now reports success as false with status 400, and its data still lists every item's result, so update any client code that treats success as false as having no results.

2026.09.26 Projects > Files Moved Out Of A Project Leave Its Agent's Knowledge

Moving a file out of a project into another folder or project now removes it from the first project agent's knowledge, just as moving a file into a project adds it to that project's agent.

2026.09.26 Files > Moving, Copying And Deleting Many Files At Once Is Much Faster

Selecting a batch of files and moving, copying or deleting them now completes as one operation instead of one file at a time. Moving 50 files takes about a tenth of a second instead of around nine seconds, and copying 50 takes well under a second instead of around sixteen. Hardened the same operations so two moves running at the same time, for example from two open tabs, both finish reliably, and other devices see deleted files disappear from pages as well as folders.

2026.09.26 Desktop > Sapience Toolbar For Quick Actions From Anywhere

The desktop app now shows a small Sapience S at the edge of your screen. Click it for a short menu of New Chat, New Note, Dictate, New Project and Search, each with its keyboard shortcut shown on hover so you can learn them as you go. Every item runs the same action as its shortcut. You can show or hide the toolbar from the Sapience tray menu, and the app remembers your choice.

2026.09.26 Projects > New Projects Set Up Their Daily Agent Job And Digest

The last step of Create New Project now has two options, both on by default: a daily Project Agent job at 9 am that works on the tasks assigned to the project's agent and emails you if it did any work, and a daily email summary of the project's activity. Both use the same setup as the Project Agent Job and Project Digest Email cards in the agent's settings, so you can change or turn them off there later. If either can't be set up, the project is still created and Sapience tells you which part to finish by hand.

2026.09.26 Projects > Deleting Projects That Were Already Removed Clears Them From The List

Deleting a project that had already been removed, for example in another window or on another device, was refused without a clear reason and left the project showing in the grid and card views. Sapience now checks with the server and, once it confirms the project is gone, removes it from the list and says it was already gone rather than deleted. A project that still exists stays in place and shows the server's reason, and a bulk delete ends in one summary instead of a notification per project.

2026.09.26 Notes > Toggle Shortcut Only Fires At The Start Of A Line

Typing > followed by a space turned the whole line into a collapsible toggle even in the middle of text, so an arrow like --> became a toggle. The shortcut now works only when > is the first character on the line, and text such as --> or a > b stays as you typed it.

2026.09.26 Agents > Plans Run Every Step And End On Their Current Status

After you approve a plan, the agent now carries out every step in the same reply, marking each step done as it finishes, rather than pausing after each step to ask whether to continue. It stops early only when it needs something from you or a step is blocked, and it can close out a plan you decide to abandon. A plan is marked complete only once all of its steps are done, and the Plan panel ends every reply showing the plan's current status and step progress. The step list is also kept when you reload the conversation.

2026.09.26 Search > Search Now Opens On Your Most-Used Items

Press Alt+F (or click Search) with nothing typed and Sapience now shows a new Frequent tab: the notes, chats, projects, files and tasks you open most, with the ones you use lately ranked highest. Your Frequent list follows your account, so it is the same in the browser and the desktop app. Start typing and search works exactly as before.

2026.09.26 Models > Chat Model Picker's Claude Opus Choice Is Now Opus 5.5

The model picker under the chat box now lists Claude Opus 5.5 as its Claude Opus option. Opus 5 is still available everywhere else you pick a model: the create-agent wizard, Advanced options, and the AI Store model browser.

2026.09.26 Agents > Choose How An Agent Uses A Spreadsheet In Its Knowledge Base

Adding a spreadsheet (csv, xls or xlsx) to an agent's Knowledge Base, when you create the agent or edit it later, now asks whether the agent should analyze it with Code Interpreter or treat it as an ordinary knowledge document. Answer Yes and the file is added for Code Interpreter; answer No and it is searched and referenced like any other document. One answer covers every spreadsheet you add in that step, other file types are added as knowledge documents as before, and Regenerate Knowledge Library on a project is unchanged. For the agent to run code on those files, switch on Code Interpreter in the agent's settings.

2026.09.26 Chat > Choose Code Interpreter Or Document When You Attach A Spreadsheet

Attaching a spreadsheet (csv, xls or xlsx) to a chat sent it to Code Interpreter automatically, with no way to have the agent simply read it. Sapience now asks each time: Yes, analyze it with the Sapience Coding Agent, or No, treat it as a regular document that the agent reads in full. The question appears on the paperclip, when you attach before starting a chat, in the upload dialog, and when you drop files onto an open chat, where it is asked once for the whole drop. Cancelling a drop attaches nothing, and files that are not spreadsheets attach exactly as before.

2026.09.26 Notes > Write With AI Works Every Time In A Session

Using Write with AI a second time in the same session, with no chat open, stopped with an AI Writing Error. Write with AI now picks up where the first use left off, so you can call it as often as you like while you edit.

2026.09.26 Chat > Deleting An Already Removed Conversation Completes Quietly

Deleting a conversation that had already been removed, for example by deleting twice in quick succession from a project's conversation list, reported an error. The delete now completes quietly and the conversation disappears from the list, while any other delete problem is still reported.

2026.09.26 Meetings > Notetakers Follow Rescheduled Meetings

Hardened how the meeting notetaker follows a meeting that is moved after it was scheduled, including a single occurrence of a recurring Outlook series. The notetaker is re-sent for the new time against the same meeting record, a move made while the meeting is already under way leaves the notetaker in the call undisturbed, and a failed re-send is retried on the next calendar sync. Meeting health reporting now also records why a calendar sync scheduled nothing for a meeting, such as the event having been deleted in the calendar.

2026.09.25 Accounts > Suspend, Cancel Or Re-Activate An Account Through The API

POST /api/admin/modify-user-status sets an account's status to active, suspended or cancelled, taking a username and new_status. A super user can change any account, an org admin can change accounts in their own organization at or below their own role, and any user can suspend or cancel their own account. Re-activating an account is reserved for an org admin or super user. A suspended or cancelled account can no longer sign in, and every change raises an account suspended, cancelled or re-activated event you can act on.

2026.09.25 API > Outbound Webhooks For New Users

You can now register webhooks that Sapience calls the moment an event happens, starting with a new user being created. Four new endpoints let you register, list and delete webhooks and see which events are eligible: POST /api/v2/webhooks/create, GET /api/v2/webhooks/list, DELETE /api/v2/webhooks/{webhook_uid} and GET /api/v2/webhooks/eligible-events. They are also available in the TypeScript SDK as client.v2.webhooks. Each webhook calls your https URL by POST or GET. It can authenticate with a bearer token, an X-API-Key header, basic auth, or nothing. Every call is signed with HMAC-SHA256 using a per-webhook secret that is shown once when you create the webhook. Organisation admins can create webhooks for their organisation. Asking for an event that isn't eligible returns success false with the code WEBHOOK_EVENT_NOT_ELIGIBLE. See the help article Outbound Webhooks From Sapience for examples.

2026.09.25 Skills > View And Edit Instructions On Skills You Own

Opening a skill you created for editing now shows its instructions field, so you can read and update how it tells the AI to execute. Instructions on a skill you don't own, including Sapience's built-in skills, stay private and are never shown.

2026.09.25 Skills > New Skills Start Active

Creating a skill now defaults its status to Active instead of Draft, so a newly created skill is usable by your agents right away without an extra step to turn it on.

2026.09.25 Chat > New Project Link In The Add To Project Window

Choosing New Project in the Add to Project window now saves the chat into the project it creates. Before, finishing the wizard also counted as a cancel, so the project was created and the picker re-opened without the chat being saved.

2026.09.25 Agents > Agent Names And Descriptions Follow Your Language

With the app set to a language other than English, agent pickers, sidebars, store cards and agent dropdowns show each agent's translated name and description where one exists. Opening the edit form loads the English original, so saving without edits sends English.

2026.09.25 Accounts > Try For Free Opens The Standard Trial Signup

Try For Free on a card in the public AI Catalogue now opens the standard free-trial signup page, and still takes you into a conversation with the agent whose card you clicked.

2026.09.25 Chat > Conversation Files List Agent Files With An Added By Column

View Files in Conversation now lists the files an agent created during the conversation as well as the ones you uploaded. A new sortable, searchable Added by column shows whether each file came from you or from an agent.

2026.09.25 Skills > Skill Owners Can Edit Their Own Instructions

In the Edit Skill window, the owner of a personal skill, and the owner or an admin of an organization skill, can view and edit the skill's instructions. The Team scope is no longer offered when creating or editing a skill, and a new skill starts as Active.

2026.09.25 Files > Files With Unreadable Pictures Show As Partially Processed

When a file's text was read but some of its pictures could not be, the file card and the reprocess progress now show it as partially processed, with how many pictures could not be read, instead of waiting until it times out. The file's AI summary is regenerated as for a completed reprocess.

2026.09.25 Files > Copy A Folder To Another Project

Copying a folder to another project now works and copies the folder with everything in it. Before, it was refused with a message that copying a folder is not supported. The destination project's file list updates straight away, and if the server refuses a copy you now see its own reason.

2026.09.25 Chat > Copy Just The Highlighted Text From A Message

Right-clicking inside a message while part of its text is highlighted now offers Copy selected, which copies exactly the highlighted text. Before, the right-click menu always copied the whole message. Copying the whole message is still available in the same menu.

2026.09.25 Mobile > Take A Photo And Attach It To Conversations

Photos taken with the camera attach item are now treated as images even when the capture has no file extension, and are sent to the agent with their true image type rather than as documents. The Camera item also shows on touch devices such as landscape phones and tablets, not only on narrow screens.

2026.09.25 Models > OpenAI GPT-6 Sol And GPT-6 Luna

OpenAI GPT-6 Sol and GPT-6 Luna can now be selected in the chat model chip, the create-agent wizard, Advanced options and the AI Store. Both read images, offer a reasoning effort setting and carry a 1,050,000-token context window. Sol defaults to high reasoning for complex coding and agent work, and Luna to low for efficient everyday use.

2026.09.25 Skills > Typed Skills Keep Your Text And Can Chain

Completing a typed /skill name with Tab or Enter no longer deletes the text around it: typing /br followed by Tab leaves /brainstorming with a space after it. You can type more than one /skill name in a message and each recognised skill shows its own badge in the order typed; the message is sent exactly as typed.

2026.09.24 Menu > Chat Sidebar Appears When Browser Storage Is Full

The app now starts normally when your browser's storage for Sapience is full. Before, start-up could stop part way, so the dashboard opened with no chat sidebar.

2026.09.24 Chat > Slow Conversation Summaries Are Not Re-Run

Hardened conversation summaries on slow connections. A second request while one is running now joins the first, a summary is given up to 90 seconds with no automatic retry, and a failed attempt is no longer re-fired by the next message or navigation.

2026.09.24 Projects > Project Files Load Once On Slow Connections

Hardened how projects load their file trees on slow connections. Identical requests already in flight are shared, and after a failure the app waits longer, up to a minute, before asking again.

2026.09.24 Projects > Assign Goals And Sub-Goals Inline

In the Tasks & Goals Manager you can now change who a goal or sub-goal is assigned to straight from its Assigned To cell, as tasks already allowed; before, clicking it opened the goal dialog. The list offers Unassigned, Me, the Project Agent and project members, and the task and goal dialogs use the same list. Only the project owner or an admin can change goal assignments.

2026.09.24 Projects > Tasks And Goals Remember Your Natural Sort

The Tasks & Goals Manager now remembers the sort you chose for each project, for each user, across refresh and navigation. Task titles sort in natural order at both goal and task level, so 1, 2, 4, 10 and Goal 1 to Goal 4 come out in sequence and 1.a comes before 1.b, instead of plain text order.

2026.09.23 Menu > Help Button No Longer Covers The Home Toggle

On Home, the floating Help button no longer sits on top of the Dashboard and Home toggle in the top-right corner. The page leaves room for it at every window width, including with longer translated labels.

2026.09.22 Chat > Hardened Long Conversation Compacting

Hardened how a long conversation is compacted once it grows past what a model can hold in a single turn. The recent stretch that Sapience keeps word-for-word is now measured in messages rather than in processing steps, so a conversation full of tool work keeps as much recent history as a plain one. When that recent stretch is still too large, oversized tool results inside it are shortened, start and end kept, before any whole exchange is moved out of it, and anything that does move out is folded into the running summary in the same pass. Your own messages and the agent's replies are never shortened, so what you asked and what the agent answered stay intact as a conversation grows. In a long, tool-heavy conversation you may notice older tool results appearing shortened where they were previously carried in full, which is what frees the room to keep more of the recent conversation exactly as it happened.

2026.09.25 Files > Creating Two Same-Name Folders At Once Gives A Clear Message

When two folders with the same name were created in the same place at the same moment, for example by an agent working on several steps in parallel, the second request could return a generic internal error. It now returns the same clear message as any other duplicate: a folder with that name already exists in that location.

2026.09.24 API > File Preview Explains An Empty File

POST /api/v2/utilities/file-to-html on a file with no contents returned a generic internal error. It now returns a 400 status code in the response body with a message saying the file is empty (0 bytes) and cannot be previewed, the same client-error response the endpoint already gives for an unsupported file type.

2026.09.23 Files > Search Stays In Your Folder By Default, With A Checkbox To Go Wider

Searching for a file used to match against every detail of every file in your account, including who owned it, which could surface files that only shared a search term with an unrelated person's email address. Search now matches only file and folder names and tags, and starts scoped to the folder you're searching from. A Search All Files/Folders checkbox sits directly under the search box, always visible, so you can widen a search to your whole account whenever you want.

2026.09.23 Menu > Sidebar Search Opens The Search Window Right Away

Switching the sidebar's search toggle to Sapience, or clicking into the search box while it is already set to Sapience, now opens the search window immediately instead of waiting for you to type and submit a query.

2026.09.23 Sharing > Hardened Rapid Share Updates

Sharing a project with several people in quick succession could occasionally report a failure even though the share itself had gone through. Sharing now retries automatically when that happens, so back-to-back share changes complete reliably.

2026.09.22 Menu > Conversation Actions Menu Stays Put

Opening a conversation's actions menu from the sidebar could show it correctly for a moment and then snap to the top-left corner of the screen. The menu now stays exactly where it opened.

2026.09.22 Chat > A Smaller, Collapsed Note When A Conversation Is Renamed Or Summarized

When Sapience automatically renames a conversation or writes a short summary of it, that notice now appears as a small, collapsed line you can expand to read, instead of a full-size chat message.

2026.09.22 Accounts > Signed-Out Sessions Return You To Sign In

When a session had ended, some screens kept quietly checking your account status in the background instead of taking you back to sign in. Sapience now recognizes an ended session right away and returns you to the sign-in screen.

2026.09.22 Meetings > Hardened Disconnected Calendar Handling

A calendar connection that had been disconnected could still be checked on every hourly sync, occasionally reporting the same meeting-join problem more than once. Sapience now recognizes a disconnected calendar consistently across every sync path and stops retrying it, so a connection problem is reported once rather than repeatedly.

2026.09.22 Files > Clearer Message When A File Has No Text To Summarize

Asking for an AI summary of a file with no extractable text, for example an image-only scan, showed a generic failure message. It now tells you plainly that the file has no text to summarize, instead of reporting an unexpected error.

2026.09.22 Files > Hardened Simultaneous Upload Pieces

Uploading a very large file split into many pieces at once could occasionally fail partway through the transfer. Sapience now handles multiple pieces of the same upload arriving together more reliably, so large uploads complete without interruption.

2026.09.22 Agents > A Clear Outcome When An Agent Reaches Its Step Limit

An agent run that reached its maximum number of steps could end marked as failed with no explanation, even when it had already produced real output like files. It now reports plainly that it reached its step limit, and lists what it had already produced before stopping.

2026.09.22 Models > Grok 4.7

Grok 4.7 is available everywhere you choose a model: the model selector in chat, the create-agent wizard, advanced conversation options, and the Sapience AI Store. It offers a 500,000-token context window, reads images, and supports the same reasoning-effort control as its predecessors, at a lower price per token.

2026.09.23 Models > Anthropic Claude Opus 5.5

Anthropic's Claude Opus 5.5 is selectable in the chat model picker, the create-agent wizard, Advanced options, and the AI Store model browser. It carries a one-million-token context window, up to 128,000 tokens of output, and reads images. Pricing is $4 per million input tokens and $20 per million output, both below Opus 5, with cached input at $0.20 per million. Two details are worth knowing if you build against it directly: unlike Opus 5, reasoning cannot be turned off at any effort level, and when no effort is specified it reasons at Medium rather than the High every other Claude model defaults to. Opus 5 remains selectable.

2026.09.22 Models > xAI Grok 4.7

xAI's Grok 4.7 is now selectable for any agent, everywhere you pick a model — the chat model picker, the create-agent wizard, Advanced options, and the AI Store model browser. It carries a 500,000-token context window with up to 450,000 tokens of output, reads images, and still takes a temperature setting for how closely it sticks to the obvious answer. Reasoning is always on across four effort levels — low, medium, high and extra high — with no way to turn it off. Pricing is $1.60 per million input tokens and $4.80 per million output, with cached input at $0.40 per million; a request whose prompt runs past 200,000 tokens is billed at double every one of those rates for the whole request, not just the part above the threshold. Grok 4.6 remains selectable if you would rather pin the earlier version.

2026.09.21 Projects > Open A Project Without Leaving Your Page

Clicking a project in Files or on your Dashboard used to take you away from whatever you were doing: Files switched over to that project's contents, and the Dashboard moved you to the Projects screen. Both now open the project in a large window over your current page, with its full dashboard inside, so Overview, Goals & Tasks, Conversations, Files & Notes, Project Team and Agent Settings all work exactly as they do on the Projects screen. Moving between those tabs leaves the page behind the window untouched, so closing it puts you back precisely where you were. When you do want the whole screen, the window's header carries a Go to Project fullscreen link that opens the project's own page on the tab you were already reading, and your browser's Back button brings you straight back.

2026.09.21 Accounts > Messages Send As The Right Account

If you are signed in to more than one Sapience account in the same browser, a message to an agent is now sent as the account whose conversation you are in, so it is no longer refused after a second sign-in.

2026.09.21 Projects > Notes In Project Subfolders Show Translations

Notes inside a subfolder of a project now switch to their translated view when you view the project in another language, as notes at the project's top level already did.

2026.09.20 Chat > Opening A Conversation That No Longer Exists

Opening a conversation that has been deleted, or that is no longer shared with you, showed its "couldn't load" notice twice. It now shows the notice once.

2026.09.20 API > Web Search Rejects An Empty Query

POST /api/v2/utilities/web-search now rejects an empty or whitespace-only query straight away, returning success false with an INVALID_REQUEST status code in the response body, instead of attempting a search that could not return results.

2026.09.20 Projects > Organization Projects Show Under Every Projects Filter

Projects shared with your whole organization now appear under every filter on the Projects page, including Active and Shared with me, and the list view has a working filter dropdown that matches the card view. A new Scope column marks them with an ORG chip, which also appears on the project's dashboard beside its status. Opening the list view from a direct link now loads your projects straight away, and the status filter sits in the grid's toolbar under a Filter label.

2026.09.20 Chat > Conversations Are Summarized After Five Messages

A conversation is now automatically summarized and renamed after five of your messages instead of three.

2026.09.20 Chat > Chat Cards Survive Updates And Reappear

After Sapience is updated while a tab is open, opening a chat card no longer shows a raw red error banner; the app retries and reloads once. Choosing a chat that is already open now brings its minimised card back instead of only saying Chat is already open.

2026.09.20 Menu > Favorite Stars On Projects, Notes, Agents And Files

A favorite star now sits beside the refresh icon on a project's dashboard, in a note's toolbar and in an agent's chat header; clicking it adds or removes the favorite and every favorites list updates. On a file's info card, the Favorite button now shows whether the file is already a favorite and toggles it.

2026.09.20 Chat > Quick Links On The Home Screen

Home now shows the dashboard's Quick Links box pinned to the bottom of the page, with its upload and new meeting actions.

2026.09.20 Files > Moving A File Already There Shows A Notice

Moving a file into a project it is already in now shows a plain notice that it is already there, in place of a red Failed to add files error.

2026.09.20 Skills > Editing A Skill From Manage Skills Saves Reliably

Hardened saving edits to a skill from Menu, Manage, Skills, so an edited skill is validated and saved consistently.

2026.09.19 Settings > One Switch Between Your Dashboard And Home Screen

Moving between the Dashboard and the Home screen used to take two different links, a "Home Screen" link on the Dashboard and a "Full Dashboard" link on Home. Both are replaced by a single Dashboard / Home switch at the top of each screen, showing the one you are on. Flip it to move to the other, and Sapience remembers your choice on that device: with Home chosen, opening the Dashboard from the sidebar, the logo or a link takes you to Home, and with Dashboard chosen, opening Home takes you to the Dashboard. If you already start on Home when you sign in, that becomes your remembered choice. Links that explicitly ask for the full Dashboard still open it, and the phone layout is unchanged.

2026.09.19 Sharing > Unpublishing Works When The Public Link Is Already Off

Unpublishing a note or project whose public link had already been turned off, for example from another tab or device or by a quick double-click, showed an "Unpublish failed" message even though the page was no longer published. Unpublish now completes normally in that case and the page shows as unpublished, while any other problem turning off a link is still reported.

2026.09.19 Notes > Notes Open From Changes Only And Attachments Show One Card

Opening Notes now fetches only the notes changed since your last visit instead of downloading every note again, and the list still shows all of them. After a hard reload, a file attached to a message shows one card instead of a duplicate on the announcement row.

2026.09.19 Chat > Start A Team Chat From The Agent Browser

The Browse AI Agents window has a Want a Team of Agents instead link that starts a team chat, and the Home agent picker's footer has a Team Chat link.

2026.09.18 Notes > The Comment Button In The Formatting Bar Now Says "Comment"

The comment action in the pop-up formatting bar that appears when you select text was a plain icon that did not say what it did. It now reads "Comment", so adding a note to a passage you have highlighted is easy to find.

2026.09.18 Agents > Longer conversations and larger messages on Grok 4.5 and Grok 4.6

Grok 4.5 and Grok 4.6 now hold substantially more context before Sapience compacts a conversation, so long working sessions with these models keep more of their history intact. The largest single message you can send to a Grok agent has also increased by 50%, which makes pasting a big document or transcript straight into the chat far less likely to be refused. When a very long conversation does need compacting, more of the recent detail now survives than before.

2026.09.18 Projects > New Notes Appear In Add Related Resource

A note created from a task's Add Related Resource window now appears in the list straight away. Before, it was created but only showed after a browser reload.

2026.09.18 Chat > Home Waits For You To Choose An Agent

Home now opens with no agent selected, and Send asks you to choose an agent first with the message Choose an agent above to start chatting. Once you choose, Home shows the agent's description, with a more link when it is cut off, its welcome message under a Welcome Message caption, and a dropdown of starter questions you can click when you click into the message box.

2026.09.18 Projects > Edit Project Translations Per Language

Project owners and project admins can now edit a project's translations for each language: the project name, goal and description, task and goal names, note content and file display names. Switching the language you view a project in reloads it in that language, and deleting or moving items warns that it affects every language. The warning shown before you change the structure of a translated project template now says it changes the template for all users and all languages, and that projects already created from it are unchanged.

2026.09.17 Notes > Trash Icon In The Comments Preview Deletes The Comment

In the comments preview beside a note's table of contents and on a file's info card, clicking a comment's trash icon now deletes that comment. Before, it opened the full comments window instead.

2026.09.17 Voice > Microphone Stays Available While You Type

The microphone in the chat composer now stays visible next to the send button while you type, and dictated words are inserted where your cursor is instead of only at the end of the message.

2026.09.17 Chat > Home Agent Picker Gains Search, Projects And Attachments

On Home, the agent picker has a search box that filters agents by name and a favorite star on the right of each row. A Show Projects and Show Agents toggle in its footer switches the list to your active projects, each addressed through its project agent. Hovering an agent previews its description beside the list, and the paperclip works: files and images you attach are carried into the new conversation with your first message.

2026.09.17 Mobile > What's New And Office Hours On Phones

The phone menu now has two extra entries below your conversations: What's New, which opens the releases page, and Office Hours, which opens the Office Hours page.

2026.09.17 Menu > Collapsed Sidebar Shows The Sapience Mark

When the sidebar is collapsed to its narrow strip, its top icon is now the Sapience S mark, a white S on black, in place of a generic sparkle icon.

2026.09.16 Menu > Faster Favorites Loading

Favorites appear promptly when you open Sapience and refresh to reflect your latest saved choices. Adding, removing and reordering favorites keeps the sidebar, agent browser and project lists consistent. Agent favorite buttons also report when a change cannot be saved.

2026.09.16 Files > Copy Files And Notes Where You Work

Right-click a file or note in Files or a project's Files & Notes tab and choose Copy (duplicate) to create a copy alongside the original, including inside nested folders. Both menus place Copy immediately before Move and Rename. You can also select files and notes together and use Actions > Copy to copy them to another folder, project or Home. The destination dialog now says Copy Here, and copy labels are available in every supported language.

2026.09.16 Notes > Open External Pages Fullscreen

Pages that show an external web page or embedded content now have a Fullscreen link above them. It opens the page in a large window that closes with its close button or Escape.

2026.09.16 Chat > A Home Screen To Ask Anything

Home is a start page for asking anything: a greeting, an agent dropdown with your favorites first, and the full chat input with its Thinking, Skills, Apps and Model selectors. Sending opens a conversation with the agent you chose. Signing in on the web with Focus Mode on by default now lands on Home, the agent picker has a View All Agents link, and the dashboard has a Home Screen link with Full Dashboard at the top right.

2026.09.16 Menu > Choose The Platform Tour Or The Power User Tour

The overview dialog now has a Tour dropdown for every user: the Platform Tour, a three-step tour that is the default, or the Power User Tour, a four-step tour. Your choice is remembered.

2026.09.15 Projects > Tasks Tab Remembers Your Last View

The Views selector on a project's Goals & Tasks tab always reset to Task List, even after you switched to Kanban, Calendar, or another view. It now remembers the view you last chose for each project — Task List, Calendar, Eisenhower Matrix, Kanban Board, Timeline, or Spreadsheet — and shows it again the next time you open that project's tasks, with no effect on any other project.

2026.09.15 Menu > Conversations Are Grouped By Kind

Your conversation history in the sidebar is now grouped into Favorites, Scheduled Jobs, Shared With Me, and Recent, and each group opens and closes on its own. Favorites and Recent stay open by default, so what you check most often is visible right away, while Scheduled Jobs and Shared With Me start folded down to keep the list short. Which group a favorited conversation appears in is now the signal that it's a favorite, so its row no longer also carries a separate star. Every row's icons for shared, scheduled, and file-carrying conversations now sit to the right of its name instead of the left, so titles read cleanly and use the row's full width.

2026.09.15 Files > Hardened Document Text Extraction

Hardened the path that pulls text out of stored documents, including PDFs, with more reliable detection of a document's true format before it is read. This broadens the range of files that can be processed, so more of what a document contains reaches search, agents and file chat.

2026.09.15 Chat > Clearer Errors When A Message Cannot Be Sent

When a chat request was declined as the reply was starting, the app described it as a stream that had closed unexpectedly, which named neither what had happened nor anything you could act on. Chat now surfaces the actual reason the request was declined, and a request that was declined outright is no longer quietly tried again.

2026.09.15 Menu > Remove Several Favorites At Once

The View All Favorites window now lets you tick several favorites, choose Select all or Clear selection, and remove the selected ones in one step after a confirmation. Before, favorites could only be removed one at a time. The Select all and Remove selected buttons match the size of the favorite names beside them.

2026.09.15 Projects > Views Selector Stays Visible On All My Tasks

The Views selector on All My Tasks, which switches between the list, matrix, kanban and other views, now stays visible when the page holds a lot of content.

2026.09.15 Mobile > Files New Menu Works On A Phone

On a phone held upright, Actions then New in Files & Notes now shows its four choices as a readable list. Before, the New fly-out was squeezed into a narrow sliver with truncated labels overlapping the toolbar.

2026.09.15 Menu > Open Sidebar Items In A New Tab

Right-clicking the Sapience logo, a conversation or a project in the sidebar now offers Open Link in New Tab, and middle-click or Ctrl-click opens it in a new tab. A plain left-click behaves as before.

2026.09.15 Voice > Dictation And Voice Prompts Spell Sapience Correctly

Voice prompts in chat and dictation into notes now spell Sapience correctly where the transcript used to read Sapiens. The scientific term Homo sapiens is left unchanged.

2026.09.14 Agents > Answers Cover Every Part Of A Multi-Part Question

Agents are guided to lead with what they found rather than build up to it. That guidance now draws a line between the order of a reply and its length: leading with the answer is about what comes first, not about how much is said, so where you ask several things at once every one of them is answered before the reply ends. Being brief applies to the wording, never to the coverage. The Project Agent template carries the same wording, so agents working inside a project follow it too.

2026.09.14 Menu > Deleted Notes Leave Your Favorites

Deleting a note now removes it from your Favorites list, so the list no longer shows notes that do not exist any more.

2026.09.14 Mobile > Tapping A Favorite On A Phone Opens It

Hardened favorites on phones. Tapping a favorite in the sidebar now always reveals the item, including when it points at the page you are already on, and closes the full-screen menu instead of leaving it open over the app.

2026.09.14 Sharing > Sharing Again Shows A Notice, Not An Error

Sharing a project again with someone who already has access now shows a short notice saying they already have access and refreshes the Team list. Before, the Team tab showed a long block of technical text as a red error.

2026.09.14 Menu > Sidebar Menus Stay Usable After Collapsing

Menus opened from the sidebar, such as a conversation's settings or Remove from Favorites, now stay usable when you have collapsed the side panel and hover to peek it open. Before, the panel could close underneath an open menu and leave it unclickable.

2026.09.13 Agents > Every Agent Picker Lists Agents The Same Way

The agent picker in the AI writer listed agents in whatever order they arrived, which did not match the sidebar. Every place you choose an agent now presents the same list in the same order: the Sapience Agent first, then the Sapience Web Research Agent, then your favourites alphabetically, then everything else alphabetically.

2026.09.13 Notes > Choose Your Diagram Editor Every Time You Draw

Notes can draw diagrams with either of two editors — the standard one, or advanced diagrams and drawing — and the advanced one was only reachable through a setting few people ever found. Adding a diagram to a note now asks which you would rather use for that diagram, every time, since the two suit different jobs; your last answer is pre-selected, so staying with one of them is a single keystroke. A diagram you have already drawn always reopens in the editor that created it, and is never asked about.

2026.09.13 Notes > Write With Agents From The Insert Menu

The AI writer in a note was reached by typing @ in the body, which is easy to miss unless somebody tells you. Insert now carries Write With Agents at the top of the menu, and typing /agents in the note opens the same panel. The @ shortcut still works exactly as before.

2026.09.13 Notes > Type A Slash To Insert Anything In A Note

Type / at the start of a line, or after a space, while editing a note and the Insert commands open right where you are typing. Keep typing to narrow them — /whiteboard or /diagram both find the whiteboard, /table finds a table — then use the arrow keys and Enter to place it, or Escape to carry on writing. The menu offers exactly what the Insert menu offers, so the two can never drift apart, and a slash inside a word or a web address is left alone.

2026.09.13 Notes > A Blank Note Starts Empty

A new blank note opened with the words Replace this with your notes already written into it, which had to be selected and deleted before you could type, and stayed in the note forever if you did not. The note now opens genuinely empty, with that line shown as grey placeholder text that disappears the moment you click in. The Table of Contents hint beside the note also now reads on a single line.

2026.09.12 Notes > Comment On A Sentence Or Paragraph, Not Just The Whole Note

A comment used to belong to a whole note. Select any stretch of text while editing one and the pop-up formatting bar now offers Comment, which attaches what you write to that exact passage. The passage carries a faint highlight from then on, wherever the note is read, and pointing at it brings the comment up without leaving the page — in the editor and in the reader alike. The comment joins the note's ordinary thread, so it still appears in the comments panel, in All Conversations and on your dashboard like any other.

2026.09.12 Notes > The Table Of Contents Panel Now Carries The Note's Comments

The floating Table of Contents beside a note now holds two sections: the contents themselves, and the note's comments. Each has a heading bar you can click to fold that section away, so you can keep whichever one you are working with. The comments section shows how many comments the note carries and the most recent of them, with New and Expand beside the heading, both of which open the full thread. The panel also no longer disappears on a note without headings — it stays, and says that adding headings will fill the contents list.

2026.09.12 Files > The Comments Area Says So When There Are No Comments

A file's details panel always kept its comments area on screen, but left it empty when nothing had been written yet, which read as though it were still loading. It now says plainly that there are no comments on the file, and carries New and Expand links that open the thread. Notes say the same thing in the same place in their Table of Contents panel.

2026.09.11 Files > Charts And Diagrams Drawn As Vector Graphics Are Now Read

Many reports and slide decks draw their charts, scores and diagrams as vector graphics rather than embedding pictures, and some presentations carry SVG images with no ordinary picture alongside them. Sapience now renders pages whose content is vector artwork, and converts SVG images inside PowerPoint, Word and Excel files, before describing them, so the figures and labels in those charts become text that search and agents can use. A rendered page is marked as the page itself rather than as a picture placed on it, and a document drawn this way is no longer described as a scanned document.

2026.09.11 Files > Sharper Image Descriptions In Files And Notes

Images in files and notes are now described by a stronger vision model, giving more accurate and complete descriptions of charts, diagrams and screenshots and more consistent results on image-heavy documents. Near-blank images such as dividers and decorative rules are recognised and noted as blank rather than described.

2026.09.11 API > A Parent Of Root Places New Items At The Top Level

Creating a folder, page, task, goal or project with `parent_uid` set to `root` now places the new item at the top level of your drive, matching how file uploads already treat `root`.

2026.09.11 Chat > The Message Box Stays In View While You Scroll Back

In the full-screen chat on larger screens, the message box now stays pinned to the bottom of the window while you scroll back through earlier messages, instead of scrolling out of view with them.

2026.09.11 Notes > Resize Images In A Note On A Touchscreen

Images attached to a note can now be resized on a touchscreen by dragging their handles, as well as with a mouse or pen.

2026.09.11 Menu > The Collapsed Sidebar Opens When You Point At It

When the sidebar is folded down to its narrow strip of icons, in Focus Mode or after you collapse it yourself, moving your pointer onto the strip now opens the full sidebar over the page, and moving away folds it back. Nothing underneath moves or resizes, so the conversation you are reading stays exactly where it was. A short pause before it opens means a pointer just passing across the strip on its way somewhere else leaves it alone. The icons line up with the sections they stand for, so the section you were reaching for is right there in the full list. Click anywhere inside the open sidebar to keep it open, which is what you want when working with one of its menus.

2026.09.11 Chat > Conversation Summaries Display As Formatted Text

Opening a conversation's Summary from the chat header now shows it in the same reader used for an agent's plan, so its headings, bold text and lists display as formatted text rather than as the symbols used to write them.

2026.09.11 Projects > Schedule A Task From Its Details, Not Just The Timeline

Setting when a task is planned to start and finish meant finding its bar on the project Timeline and right-clicking it. Open any task's details and you will now find Advanced Scheduling inside Additional Notes, with a Start Date and a Planned Finish, the task's due date shown alongside them for reference, and a running count of how many days the plan covers and how it sits against that due date. Clear schedule dates puts the task back to being drawn from when it was created through to its due date. The dates you set here are the same ones the Timeline draws, so a task scheduled from its details moves on the chart straight away, and the due date itself is never changed by scheduling it. A plan that starts or finishes after the due date is allowed on purpose — a project running late still has to be planned — and shows up as a predicted missed deadline rather than being refused.

2026.09.11 Chat > Comment Threads Appear In All Conversations

A comment thread on a note, file or task is a conversation, and it now shows up as one. The list — renamed All Conversations and Agent Sessions — carries your comment threads alongside your chats, with a new Type column marking each row as Chat or Comment. A comment row is named after the item it hangs off, so you see Comments on "The Beast Updates" note rather than an identifier, and it shows the same message count and unread count as any other row. Clicking one opens that item's comment panel, which now names the item in its header too. Mark All Read clears unread comments along with unread conversations.

2026.09.11 Chat > View Unread Now Filters The Conversation List

View Unread brought unread conversations to the top of the list by sorting on the unread column, which is hard to pick out on an account carrying thousands of rows. It now filters: press it and the list shows only what actually has something unread, chats and comment threads alike, still ordered by how much is waiting. The button reads Clear Filter while the filter is on, so the way back is always on screen, and the list opens unfiltered every time.

2026.09.11 Notes > See Your Unread Comments From The Dashboard

When someone comments on one of your notes, files or tasks, a blue chip now appears beside the Activity Feed heading on your dashboard telling you how many unread comments are waiting. It shows only when there is something to read, and clicking it opens All Conversations and Agent Sessions, where you can see which items the comments are on and open each thread.

2026.09.11 Files > Comments Bring A File Or Note Back Into Recent Files & Notes

A file or note somebody has commented on now counts as recently active. Recent Files & Notes on your dashboard ranks on the later of when the item itself last changed and when it was last commented on, so something your team has been discussing appears there even if nobody has edited it, and the date on the row reflects the comment. An item with no comments is ranked exactly as it was before.

2026.09.11 Settings > The Dashboard Activity Feed Is Always Open

The Activity Feed on your dashboard could be collapsed down to its heading, and on a phone it started out that way. It is now always open, on every device, so the feed is there as soon as the dashboard is.

2026.09.11 Views > Views And Reports Recovers From A Failed Load

Hardened how a Views & Reports screen loads the grid it runs on. If that part of the page cannot be fetched — most commonly when a tab has been left open across an update — Sapience now retries on its own, and refreshes once to pick up the current version if the retry does not settle it, rather than stopping with an error. That refresh is capped at once per session, so a screen that stays unhappy will not keep reloading itself.

2026.09.10 Files > Images Are Read Even When A Document Has No Text Layer

Image reading now covers documents that have no readable text layer, such as scanned PDFs. The pictures inside PDF, Word, PowerPoint and Excel files are described whether or not the file's own text could be extracted, and the stored content states plainly when its text came from reading those pictures.

2026.09.10 Files > Every Page And Slide Is Accounted For When Images Are Read

Hardened page and slide tracking when reading the images in a document. Each picture is now tied to the page or slide it appears on, including a picture reused across several slides, and any page that could not be read is named in the stored content, so the result always accounts for the whole document.

2026.09.10 Files > Faster Folder And Project Opens

Opening a folder or project is now noticeably faster, especially on large accounts.

2026.09.10 Chat > Hardened Conversation History For Reasoning Models

Hardened how earlier turns are replayed to reasoning models, so long conversations with those models continue reliably from one turn to the next.

2026.09.10 Files > Uploads Into A Folder With A Matching File Name Are Refused Up Front

Uploading a file into a folder that already holds a file with the same name, ignoring case, is now refused before anything is stored, with a clear conflict response the app can show you.

2026.09.10 Mobile > A Cleaner Files & Notes Layout On Mobile

On mobile, Files & Notes now uses a single-row header, stacks the search box above the folder path, lets the toolbar scroll sideways, and shows the file list edge to edge.

2026.09.10 Settings > Start In Focus Mode By Default

A new Use Focus Mode by Default setting, under Settings & Customization > Advanced, starts Sapience with the sidebar collapsed every time you open it.

2026.09.10 Projects > The Timeline View Is Available To Everyone

The Timeline view for goals and tasks in Projects has left preview and is now available to everyone, with no setting to turn on first.

2026.09.10 Projects > Assign And Set A Due Date From The Quick Add Box

Typing @ in the Quick Add box, on a project's Tasks & Goals tab or on All My Tasks, now opens a menu of people and dates. Pick a person to assign the task, or pick a date — today, tomorrow, a day of the week, "in 3 days", "sep 14" and similar — to set its due date, all without opening the full task dialog. Arrow keys move through the menu and Tab accepts whichever entry is highlighted, or the only one if there is just one match, so the whole thing works from the keyboard. Picking an entry completes what you typed right there in the box — @ken becomes @mba.ken — rather than clearing it, and pressing Enter creates the task with a clean title plus whatever you assigned and set.

2026.09.10 Projects > Delete Any Combination Of Goals And Tasks

Selecting more than one goal in the Tasks & Goals Manager left every delete option unavailable, and the quick-actions bar appeared only for a selection made up entirely of tasks. You can now select any combination — several empty goals, a goal together with its tasks, or a mix of goals and loose tasks — and delete them in one go from the quick-actions bar, which reads "N selected" alongside Delete. When a selected goal has children, a confirmation names the full blast radius first, counting every task and sub-goal that will go with it, so the number you are shown is the number that is removed. Selecting a goal automatically includes everything beneath it, and those items are deleted once as part of their goal rather than twice. The three bulk edit actions - set status, change due date and set assignee - continue to apply to task-only selections.

2026.09.10 API > Comments On Files, Notes And Tasks

You can now hold a discussion against a file, note or task through a single new verb, POST /v2/srm/comments, which takes an operation of list, get or new. Listing is keyed on the item's uid and pages through the thread oldest-first with an exclusive cursor, deliberately the opposite order to the tree endpoint, so a reader follows the discussion the way it was written; each comment comes back with its author's username, its body, when it was written, and whether the caller may delete it. Agents author comments natively alongside people, so a comment written by an agent is marked as one. Deleting a comment goes through the canonical POST /v2/srm/delete with a type of "comment" rather than a verb of its own, so a client that already deletes files and notes deletes comments the same way. Because the comment identifier is required and comments carry no uid prefix, that type must be supplied explicitly.

2026.09.10 Sharing > Comment Access Follows The Item A Comment Is Attached To

A comment thread takes its permissions from the item it hangs off, not from itself. Anyone who can read a file, note or task can comment on it, so a colleague you have given view access to a note can join the discussion on it without being given edit rights to the note itself. A comment can be removed by the person who wrote it or by the owner of the item it sits on. Because a thread has no permissions of its own to grant, it is never listed among your conversations and cannot be shared or summarized separately from the item it belongs to. Anyone with no access to the item sees an empty thread rather than an error.

2026.09.09 Notes > Your Notes And Pages Stay Consistent On Your Device

Hardened the way Sapience keeps a copy of your own notes and pages on your device, so a note is stored there consistently — including when it carries details such as its tags or its sharing settings. This covers both saving a single note and saving many at once, which now go through the same checks, so your notes stay in step between sessions and while you are offline, rather than needing a reload to come back into line.

2026.09.09 Chat > Chat Stays Usable When An Agent's Reply Cannot Be Retrieved

Hardened what happens when the background work behind an agent's reply can no longer be retrieved. The conversation now stays usable — the typing indicator stops and the message box is re-enabled — and Sapience shows a clear message with a reference code you can quote to support, rather than leaving the conversation waiting on a reply that is not coming.

2026.09.08 Notes > Faster Saves For Frequently Updated Items

Saving a new version of an item now touches only the item's current record instead of its whole version history, so items that have been updated hundreds of times save as quickly as new ones. Version history itself is unchanged and remains fully available.

2026.09.08 API > Faster Event History Reads

Reads of a user's event history filtered by event type, used by the system events endpoint and the meeting notetaker health report, are now served from a dedicated index and return in single-digit milliseconds regardless of how much history has accumulated.

2026.09.08 Projects > Consistent Visibility Rule Across Every Tree Read

The rule introduced on 7 September, where anything beneath a project or folder you own is visible to you whoever created it, now governs the older tree and list endpoints as well as the main tree call, so every read of your workspace agrees on what you can see and the heavier sharing lookup those endpoints used has been retired from them too.

2026.09.08 Agents > Project Agents Stay Inside Their Own Project

A project's agent now keeps its everyday work inside that project — it edits the notes and files that belong to the project it's running in, rather than reaching into notes, files or projects elsewhere in your account. When a project agent does need to look further afield, such as searching your other notes and files or listing your other projects, it now asks first and only goes ahead once you approve it.

2026.09.08 Sharing > Shared Project Permissions Now Match What You Can Actually Do

A project's sharing permission now decides exactly what you can do inside it, on both ends. If you can only view a project someone shared with you, neither you nor its agent can create or edit notes, files or folders in it — that now requires edit access. And if you do have edit access, you can now also edit a teammate's file inside the project, including looking at its version history, matching what you could already do with notes.

2026.09.08 Agents > Agent Search Now Matches What You Can See Everywhere Else

Search carried out on your behalf by an agent now goes through the same access checks as the rest of Sapience, so what it can find lines up with what you can see in your files, notes and projects elsewhere in the app.

2026.09.08 Connected Apps > Airtable And Asana

Airtable and Asana now have dedicated pages under Team & Enterprise, with app-specific guidance and three clickable starter questions each. Airtable offers prompts for listing bases, viewing records and adding a record; Asana offers prompts for reviewing incomplete tasks, finding upcoming deadlines and creating a task.

2026.09.08 Connected Apps > Microsoft Apps Group

Microsoft Email (Outlook), OneDrive, SharePoint, Excel and Teams now sit together in a collapsible Microsoft Apps group under Team & Enterprise in the sidebar and connection manager. Each app keeps its own connection status and controls, and opening a Microsoft app page expands its sidebar group automatically.

2026.09.08 Connected Apps > Google Calendar And Google Tasks

Google Calendar and Google Tasks now have dedicated pages under Individual > Google Apps, following Google Slides. Each page includes app-specific guidance and three clickable starter questions: Calendar offers prompts for reviewing your schedule, finding free time and adding an event; Tasks offers prompts for reviewing due or overdue tasks and creating a task. Both apps also appear in the connection manager and chat app picker.

2026.09.08 Notes > Notes Shared With You Stay Consistent On Your Device

Hardened the way Sapience keeps a copy of a note shared with you on your own device, so a note that reaches you through a share is stored there consistently, including when it carries details such as its tags. Notes shared with you now stay in step from the moment you sign in and while you are offline, rather than needing a reload to come back into line.

2026.09.08 Accounts > Your Session Is Restored At Start-Up, Not Only Once You Are Underway

Hardened how the desktop app restores your sign-in session. The quiet renewal that keeps you signed in is now ready for the very first request the app makes as it starts, where previously it became available only once the app was fully underway. A session that can be renewed without your involvement can now be renewed at start-up, rather than returning you to the sign-in screen.

2026.09.07 Agents > Hardened The Agent HTTP Request Tools

The http_get and http_post tools an Agent uses to call remote endpoints and webhooks have been rebuilt on an asynchronous HTTP client. A slow or unresponsive remote endpoint now only affects that one tool call; everything else in the workspace keeps responding normally while it waits, and the tool's response shape is unchanged.

2026.09.07 Files > Hardened Content Checks After Document Parsing

After a document is parsed, its extracted text is checked for quality before it is indexed. That check now runs off the main request path, so very large uploads no longer delay other requests while their content is validated. The check itself and its verdicts are unchanged.

2026.09.07 Sharing > Faster Share Details On Large Shared Projects

When many items sit under one shared project or folder, the sharing details attached to each item are now resolved once per share and reused across the items beneath it, instead of being rebuilt item by item. Reads of large shared trees spend less time assembling sharing information; what each viewer sees is exactly as before.

2026.09.07 API > Tree Items Say Why You Can See Them, Including Items Inside Your Own Containers

Every item returned by POST /v2/srm/tree carries a visibility value, and that vocabulary gains one word: owned_container_ancestor, for an item that someone else created inside a project, folder or page you own. The four values are now owner, org, shared and owned_container_ancestor. An item marked shared always carries its share object (who shared it, the permission level, and the container the share was made on), and a file that is org-scoped reports org rather than shared. The change is additive; a client that meets a visibility value it does not know should treat the item as shared.

2026.09.07 Sharing > Everything Inside A Container You Own Is Yours To See And Edit

A project, folder or page you own now lists everything created inside it by the people you collaborate with, whether or not anything in it has been shared out, and you can open and edit those items. The tree listing, the batch get endpoint, the item endpoints and the Browse-by-Tag vocabulary all apply the same rule, so what you see in a list is what you can act on. Listing what you can see also no longer grows with how much everyone else in your organization shares, which keeps large accounts responsive.

2026.09.07 Files > The Web App Starts With Fewer Requests

When Sapience starts, it no longer fetches the shared-with-me list, your outbound shares, or the tag vocabulary on behalf of files, notes and projects. Sharing state for those items is read from the tree listing they already arrive in, outbound shares are read when you open a note, and the tag vocabulary is loaded the first time you browse or pick tags. Conversations shared with you are still loaded at startup, since they carry the names shown in group chats.

2026.09.07 AI Models > Google Gemini 3.8 Flash

Google has released Gemini 3.8 Flash, and you can select it for any agent today. It carries a 1,048,576-token context window with up to 65,536 tokens of output, and it reads image, video, audio and PDF input alongside text. Its thinking is steerable across three levels — low, medium and high — and Sapience passes that setting through to the model, so an agent set to low answers with almost no reasoning step while one set to high thinks at length before it replies. Pricing is $1.50 per million input tokens and $7.50 per million output, with cached input at $0.15, and it is flat: there is no higher rate above a context threshold, so a long prompt costs the same per token as a short one. Because the pricing is flat and the window is large, Sapience keeps your full conversation history and file contents in context on this model rather than summarizing them as a conversation grows. Gemini 3.7 Flash remains selectable if you would rather pin the earlier version.

2026.09.06 Files > Uploading A File Into A Folder Finishes Quietly

Uploading a file directly into a folder could report that the upload had failed, even though the file uploaded correctly and was already sitting in the folder you chose. Sapience was following the upload with a request to move the file into the folder it was already in, which was rightly declined, and was treating that as a failure of the upload. The redundant step has been removed, so choosing a folder and uploading into it now completes without anything being reported.

2026.09.06 Accounts > An Expired Session Is Reported As An Expired Session

When your sign-in session expired while Sapience was saving a change to your settings, the message that came back described an unrelated technical problem rather than the real one, which made an ordinary expired session look like something had gone wrong with the request. An expired session is now recognised as exactly that and reported plainly, so it is clear that you simply need to sign in again.

2026.09.05 API > The Tree And Get Endpoints Now Do Page-Sized Work

POST /v2/srm/tree and POST /v2/srm/get now compute only the page you asked for. A list request used to gather everything you could see and cut the page afterwards, so the cost of one page grew with the size of your organization's content; it now grows with the page size alone, and content that is shared across an organization is prepared once and served to every member. Request and response shapes, cursors and page sizes are unchanged, so no client changes are needed.

2026.09.05 Sharing > A Folder Shared From Inside A Project Now Lists Everything Beneath It

When a folder nested inside a project was shared with you, its contents appeared only when you opened that folder directly; the full tree listing showed the folder on its own. Everything beneath a shared folder now appears in the full listing as well, and each item carries the permission of the folder it was shared through.

2026.09.05 Files > A File Inside A Page Reports Its Full Path

A file attached inside a page reported its path as though the page sat at the top level. It now reports the full path through the page's project and folder, matching what you already see for files kept in folders.

2026.09.05 AI Models > OpenAI GPT-6 Astra

OpenAI has released GPT-6 Astra, and you can select it for any agent today. It is their most capable model for hard, end-to-end work, with a 1,050,000-token context window, up to 128,000 tokens of output, image input, and a reasoning effort you can set from low through to max. Pricing is $10 per million input tokens and $50 per million output, with cached input at $1; a prompt carrying more than 272,000 input tokens is charged at twice the input rate and one and a half times the output rate, applied to the whole request rather than just the part above the threshold. Temperature is not a control on this model — like the rest of the GPT-6 and GPT-5.6 family it reasons instead, so Sapience does not send one.

2026.09.05 AI Models > Zhipu GLM-5.3

GLM-5.3 from Zhipu is now selectable for any agent. It carries roughly a million tokens of context and up to 128,000 tokens of output, and its reasoning is always on, steerable across three levels — low, high and max — with no middle setting. It is a text-only model, so it does not take image input. Pricing is $2.10 per million input tokens and $6.60 per million output, with cached input at $0.21. The Sapience Agent (GLM) in the AI Store now runs on 5.3, so a conversation with it picks the new model up without you choosing anything, and GLM-5.2 remains selectable if you would rather pin the earlier version.

2026.09.05 AI Models > Anthropic Claude Fable 5.1

Anthropic has released Claude Fable 5.1, and you can select it for any agent today. It is built for demanding reasoning and long-horizon agentic work, with a one-million-token context window as standard, up to 128,000 tokens of output, image input, and adaptive thinking that is always on and steered by effort, defaulting to high. Pricing matches Fable 5 at $10 per million input tokens and $50 per million output, but cache reads cost a quarter of what they do on other Claude models, at $0.25 per million — which tells on long sessions that re-read a cached prefix. Two API details are worth knowing if you build against it directly: it does not accept a forced tool choice, and it does not take a temperature. Claude Fable 5 remains selectable.

2026.09.05 Files > Your File List Stays Consistent On Your Device

Hardened the way Sapience keeps a copy of your file list on your own device, so a file you have just uploaded, moved, renamed, or had an agent create is stored there consistently — including when it carries details such as its tags or the folders it belongs to. Your files and their details now stay in step between sessions and while you are offline, rather than needing a reload to come back into line.

2026.09.04 Accounts > Signing Out Finishes Cleanly

Signing out could surface an error part-way through, even though the sign-out itself completed and you were signed out correctly. The sign-out sequence now handles the moment at which your session details have already been cleared, so it finishes quietly with nothing reported.

2026.09.04 Projects > A Deleted Task Or Goal Stays Deleted

A task or goal you deleted could reappear after a page reload, and deleting it a second time reported that it no longer existed. The delete always worked but sometimes a cache remained for five minutes. Deletes now show/happen no matter what. Sapience clients now keeps its own record of what you have removed and that record survives a reload, so an item you delete is gone from a project's Goals & Tasks tab and every view on it, from All My Tasks, from Views & Reports, from search results, and from the task and goal counts shown on project cards. Deleting a goal takes its tasks with it in the same way. Once the item is confirmed gone everywhere, Sapience stops holding the record, so nothing you create later is affected by it.

2026.09.03 Projects > A Timeline View For Goals And Tasks, In Preview

Goals and tasks can now be read as a timeline. Turn on Preview: Timeline Mode in Projects under Menu ▸ Settings ▸ Advanced, and Timeline joins the view picker both on a project's Goals & Tasks tab and in All My Tasks across every project. Tasks are drawn as bars on a date axis, grouped into lanes by project or by assignee, with goals marked as milestones on the day they are due. You can zoom by day, week or month, jump to today, drag a bar to reschedule it, fold a task's sub-tasks away, and highlight the critical path. Two things a task can now carry make this work and are available to you directly through the API and SDK as well: gantt_start_date, the day work on it is scheduled to begin, which is what a bar starts on rather than the day the task was created; and gantt_depends_on, the tasks that must finish before it, stored on the task that waits. A dependency you draw is saved with the task itself, so it is there for your teammates and on your other devices. The setting stays off until you turn it on, and turning it off puts every view back exactly as it was.

2026.09.03 Email > An Instruction About Language Now Holds On Emailed Agent Replies

Language handling on emailed agent replies is now consistent with the rest of Sapience. An instruction you have given an agent about which language to answer in is carried through to a reply it sends by email, rather than being overridden on the way out. Emailed replies are also now free of the progress lines an agent shows while it is working — those belong on screen while you watch it think, not in the message that arrives in your inbox.

2026.09.03 Email > A Deadline You Mention Lands In The Task's Due Date

Asking an agent by email to create a task with a deadline recorded that deadline in the task's text and left its due date empty, so the task did not appear where you would expect it in views, sorts and filters that work by date. A date stated in your request now goes into the due date field itself, whatever language you write in. A date that cannot be read still creates the task, without a due date rather than failing.

2026.09.03 Chat > Charts In An Answer Render Reliably As It Arrives

Hardened chart rendering in chat answers so that a reply containing one or more charts draws them consistently while the answer is still being written. Every chart in an answer is now claimed for drawing in a single pass, so a long reply that updates many times as it arrives ends with all of its charts on screen. A chart whose data genuinely cannot be read still reports that on its own, without affecting the others in the same answer.

2026.09.03 Chat > The Conversation Summary Notice Follows Your Language

When Sapience summarizes a conversation and gives it a name, it posts a short notice in the thread saying so. That notice was written in English whatever language you were using. It now appears in your own language, in all eleven languages Sapience supports.

2026.09.03 Mobile > Wide Tables In An Answer Scroll Instead Of Squeezing

A table inside a chat answer was always shrunk to fit the width of a phone screen, which broke column headings mid-word once there were more columns than could readably fit. A table wide enough that its columns can no longer be read now keeps its natural width and scrolls sideways within the answer. Narrower tables are unchanged and still fit the screen without scrolling.

2026.09.03 Mobile > The Project Picker Fits The Screen

Creating a task from the New menu on a phone opened a picker for choosing its project that was wider than the screen, so its right-hand edge and part of the project list sat out of reach. The picker now fills the width actually available to it, up to the same maximum width it has always used on a larger screen.

2026.09.03 AI Models > Meta Muse Spark 1.3

Meta has released Muse Spark 1.3, and you can select it for any agent today. Meta has tuned this checkpoint for long-running, multi-step agent workflows, with stronger coding and instruction following, and it keeps everything that made 1.2 good — fast, large context, multi-modal, and strong at agentic tool use and delegation. Pricing is unchanged from 1.2. It is also now the model behind Project agents by default, so a project you create picks it up without you choosing anything, and the Sapience Agent (Muse Spark) in the AI Store runs on it too. 1.2 and 1.1 both remain selectable if you would rather pin an earlier checkpoint.

2026.09.01 Sharing > A Shared Project Link Now Carries Its Files, Folders And Notes

The public page for a shared project showed the project's details and its goals and tasks, and nothing else — the files and notes that make up most of the work were not there. It now opens with a contents list linking to Overview, Goals & Tasks, and Files & Notes, and the Files & Notes section shows the project's folders, notes and files as a tree you can open and close, however deeply they are nested. Goals and tasks now start collapsed to their top level so a large plan fits on a screen, with Expand all and Collapse all still there. Whoever you send the link to can open a note in place, without leaving the page, and download a file straight from the list. The whole page still prints to PDF with everything expanded.

2026.09.01 Sharing > Share A Single Folder By Link

A folder can now be published as a link of its own, the same way a project or a note already could. The page shows that folder's contents — its notes, its files, and any folders inside it — and nothing above or beside it. Password, expiry and view limit work exactly as they do for every other kind of share link.

2026.09.01 Sharing > You Are Told When A Share Link Reaches Its View Limit

A share link can be set to stop working after a number of views, and until now it simply stopped, with nothing to tell you it had. Sapience now emails you once, at the moment a link of yours reaches its limit, so you know the person you sent it to can no longer open it. Browsing within a single shared project no longer eats into that count either — opening the shared page counts as one view, and opening a note or downloading a file from inside it does not.

2026.09.01 Settings > Your Dashboard Fills In All At Once And Stays Current

The Activity Feed sat on a loading message for several seconds after the rest of your dashboard had already filled in, and once it had loaded it did not change again, so a note or file you edited while looking at the dashboard only appeared after a reload. Every module now draws on what Sapience has already loaded for you, so the whole dashboard appears together and each module keeps itself up to date as newer information arrives in the background. Editing a note, uploading a file or completing a task while the dashboard is open now shows in the Activity Feed straight away. The Onboarding & Setup tile fills in the same way, and a module that is still filling in is now shown as such rather than as empty, so a section you genuinely have nothing in no longer looks the same as one that has not finished.

2026.09.01 Notes > Attach Files To A Note

A note can now hold files. Attach them from the note's Actions menu, from the editor's Insert menu, or by dragging a file straight onto the note you are reading or editing — and on a phone or tablet, by tapping the same spot to pick one. Everything attached shows as a card at the bottom of the note, with the full filename on display rather than trimmed to fit, and you can remove an attachment without deleting the file itself. Drop a file at a particular point in the note body and it is also referenced right there, alongside the list at the end. Sharing a note now shares whatever is attached to it, at the note's own permission level, and duplicating or deleting a note carries its attachments with it.

2026.09.01 Desktop > Desktop updates now take you straight to the newest version

Checking for updates on the Sapience desktop app now brings you to the very latest build in a single update, instead of stepping through intermediate versions one restart at a time. The desktop app and the download links on the website now read the same source for what the newest build is, so a manual download and an in-app update always agree. If the app genuinely cannot reach that source, it now tells you it could not check, rather than reporting that you are up to date.

2026.08.31 Agents > The EFQM Management Report Is Now A Navigable Document

An assembled EFQM Management Report is long, and reading it meant scrolling it end to end. It now carries a pinned contents rail that tracks where you are as you scroll, becoming a button and a slide-out panel on a narrow screen, and Expand all and Collapse all controls at the top of the document. EFQM's own guidance for each Sub-Criterion is open by default rather than hidden, and each applicant answer can be collapsed by clicking its own heading, so you can move between what was asked and what was written without losing your place. The masthead now names the model and labels the applicant, and the report records its date of submission. The document still prints cleanly to PDF, and reads correctly right to left in Arabic.

2026.08.31 Sharing > Shared Project Pages Read And Print Better

The public page for a shared project now shows who a task or goal is assigned to as a chip on its summary row, next to the status, so the owner is visible while the item is still collapsed — and an unassigned item says so quietly rather than looking like a second status. The small-caps labels above each section and field are set in the page's dark ink rather than a light grey, and the project's description now carries the same weight as the page's other sub-headings. Printing a large plan no longer gives a big top-level goal a page of its own with its tasks pushed to the next one; goals now flow across pages while keeping smaller nested items intact.

2026.08.31 Projects > An Earlier Version Of A Project Shows The Files It Actually Had

Opening a past version of a project listed the files the project holds now, rather than the ones it held when that version was saved, so a file added later appeared in a snapshot taken before it existed. Each version now lists its own files, and the current version is unchanged.

2026.08.31 Files > Deleting A Folder Tells You Every Reason It Cannot Be Deleted

When a folder could not be deleted for more than one reason, only the first reason found was reported, so resolving it revealed the next one instead of finishing the job. Both blockers are now named together in a single message. The rules themselves are unchanged — nothing that was protected before is deletable now.

2026.08.31 Files > Cloning A Folder Leaves Nothing Behind If It Cannot Finish

Hardened folder cloning so that a clone which stops part-way through does not leave a partly-assembled folder in your Files. A clone now either appears complete or does not appear at all, and the views that list folders are refreshed consistently once it finishes.

2026.08.31 Sharing > A Share Page Now Shows Why A Link Could Not Be Opened

When a shared link could not be opened, the page fell back to a generic message rather than the explanation it had actually been given, so the reason was lost on the way to the screen. That explanation now reaches the page, and it is shown to whoever followed the link. Every public share page — a shared note, file, conversation, message or project — is now produced by one shared page builder, so all of them present a link that cannot be opened the same way.

2026.08.31 Files > Share A Folder With Your Organization From Its Right-Click Menu

Right-clicking a folder on the Files page now offers Promote to Org Folder, so a folder can be shared with your whole organization in one action instead of first ticking its checkbox and opening the Actions menu. A folder is shared in its own right rather than through what it holds, so this works on an empty folder exactly as it does on a full one, and the existing Actions ▸ Share with org path now covers folders alongside files in the same selection and reports a single result for everything you picked. Where a folder cannot be shared — because it already is, or because someone else owns it — you are now told which of the two applies.

2026.08.31 Voice > Dictation Works On Every Screen, With Or Without A Chat Open

Hardened the dictation microphone so it behaves the same wherever it appears — task and goal descriptions, agent instructions, settings fields, note editors and the chat composer — whether or not a conversation is open. Dictation is now a self-contained control: it records, transcribes, and hands the text to the field you are typing in, without depending on anything else on the screen. Recovery from an interrupted or abandoned recording is more robust, including when a dialog is closed part-way through a recording, so the microphone stays ready for the next use. Dictation also no longer saves an audio file when you are simply entering text, so the words appear sooner and your Files stay clear of recordings you did not ask for; the note editors, where the recording itself is the point, still keep theirs.

2026.08.31 Notes > Voice Input In The Mobile Note Editor Matches The Desktop Editor

Hardened voice input in the note editor on a phone so that it and the desktop editor share a single path for handling a finished recording. Dictating into a note now behaves identically on both: the transcript is inserted where you are writing, and ticking "Also insert playable audio" adds the playable clip to the note alongside it.

2026.08.30 Sharing > Shared Projects Keep Their Badge In The All Projects View

On the Projects page with the All filter selected, a project shared with you showed its SHARED badge for a moment as the page settled and then lost it, even though the project itself stayed in the list and still read correctly under the Shared With Me filter. A project now carries one record of itself throughout the app, so its sharing state is the same fact wherever it is read: the badge appears as soon as the project does and stays for as long as it is listed, whichever filter you are on and whichever order the page happens to load in.

2026.08.30 Agents > EFQM Management Report Assembly Is Deterministic, Verbatim And Localizable

Hardened the EFQM Management Report assembler so that a folder of notes and EFQM Results Input Table workbooks produces the same, consistently formatted document every time it runs. Results indicators are now carried as structured data read straight from the workbook cells — indicator name, unit, trend direction, and the Actual, Target and Benchmark figures for each period — so every figure in the report matches the workbook it came from, and indicator slots left as supplied template text are recorded but not shown. Narrative content is reproduced exactly as written, with the author's own formatting intact. Where content appears to belong under a different Sub-Criterion, the report flags it where it was written and adds a cross-reference at the Sub-Criterion it relates to, both clearly marked as AI-authored; nothing the submitter wrote is rewritten, corrected or moved. All 29 Sub-Criterion titles follow the official EFQM 2025 model.

2026.08.30 API > Generating An EFQM Management Report Now Takes A Required Language

POST /v2/utilities/tools/execute with tool_name "generate_efqm_report" now requires a language parameter alongside folder_uid, and a call without one is rejected rather than assuming English. The value is recorded on the report and selects the language of the model's own headings, column titles and labels, while the submitted content stays in the language it was written; Arabic renders right to left. The assembled document is returned as a file as before, and running the tool again on the same folder now creates a new version of that folder's report rather than a second, unrelated one, so a submission keeps a single record with its own history.

2026.08.30 Files > Project File Lists Are More Complete, Accurate And Faster

Hardened how a project's file list is assembled so it is derived directly from where each file actually lives, rather than from a separate record kept alongside it. A project now lists exactly the files held in it and in its folders, at any nesting depth, and that list cannot fall out of step with the files themselves — including when a file is renamed, moved or removed. Each file's description and summary are read fresh, so an agent working in the project sees a file's current details rather than the ones it had when it was added, and large projects load faster because there is less to fetch on every read.

2026.08.30 Files > Moving A File Between Folders Reaches Every View Promptly

Hardened how a file move is signalled to the rest of the app so that both the folder it left and the folder it arrived in, along with the projects containing them, are marked as changed at the moment of the move. Clients that refresh incrementally rather than reloading everything now pick up the new location on their next refresh, so a file that has been moved appears in its new home and leaves its old one without a full reload.

2026.08.30 Projects > Refreshing A Project Updates The Project Itself

Refreshing a project — from the refresh button in its header, or Refresh from Cloud in its Actions menu — reloaded the project's tasks, goals, notes and files, but not the project's own details. A change made to the project elsewhere, such as publishing it, renaming it or changing its status, only appeared after a full browser reload. Refreshing now reloads the project record alongside everything inside it, in a single request, so the header shows the project's current state as soon as the refresh finishes.

2026.08.30 Projects > Published Status And Documentation Link Move To The Right

A project's Published row and its Projects documentation link sat beneath the project's description on the left-hand side of the header. Both now sit on the right, directly under the Chat and Actions buttons, and reflow below those buttons on a narrow screen. Everything they do is unchanged — the Published row still toggles a public link and offers Copy URL when one exists.

2026.08.29 Projects > Moving And Deleting Tasks Takes Effect Immediately

Dragging a task or goal onto a new parent in a project's Tasks & Goals Manager, and deleting tasks from it, both left the list unchanged for several seconds while the change was saved, with nothing in between to say it was on its way. Moving a task or goal now takes effect the moment you drop it, and deleting tasks removes them straight away, with the rest of the app staying usable while the change is saved. If a change is not accepted, the affected row comes back at exactly the position it held before, and a message names what did not go through. Deleting a goal still takes a moment to finish, because removing a goal also removes the goals and tasks beneath it. Confirmation prompts are unchanged, and every Delete control on the page — the Actions menu, the floating selection toolbar, and the task and goal detail dialogs — now behaves the same way.

2026.08.29 API > Public Link Sharing Now Supports Projects

POST /v2/sharing/create-share now accepts resource_type "project" alongside page, file, message and conversation, and returns a share_url that renders the project as a standalone public page anyone with the link can open. The page leads with the project's overview — its primary goal, success criteria, target completion date, status, owner and notes — and then its full goals and tasks breakdown as a nested tree, each item showing its status, priority and completed count, and each expanding to reveal its own detail. Expand all and Collapse all controls make a large plan quick to skim, and the page is built to print cleanly to PDF with every item opened. The password, expiry and view-limit options behave exactly as they do for a shared note. Files, conversations, notes and project members are deliberately not included on the page. This is available through the API now; the control to create one from inside the app is still to come.

2026.08.29 Projects > Custom Goal And Sub-goal Labels

Goals and sub-goals in a project's Tasks & Goals view have always carried a fixed "Goal:" / "Sub-Goal:" label. Editing a project now offers a "Re-map Goal Labels" option that lets you choose your own wording instead — call them "Phase" and "Stage", for example — or remove the label entirely. The choice is scoped to that project, so different projects can use different vocabulary, and every existing project keeps its current labels until you opt in.

2026.08.29 Projects > Tasks And Goals Appear The Moment You Add Them

Adding a task or a goal from the dialog on a project's Tasks & Goals Manager left you looking at an unchanged list: the dialog closed straight away, nothing appeared, and the new row only turned up several seconds later, with nothing in between to say it was on its way. The row now appears the instant you save, in its right place under the goal you chose. If it cannot be stored, the row is removed again and a message names the task or goal that did not save, so what you see always matches what actually exists. The Quick Add box above your task list now works exactly the same way, so adding through either route behaves identically.

2026.08.29 API > Moving And Copying Items Is Substantially Faster In Bulk

Hardened the move and copy paths behind POST /v2/srm/move and POST /v2/srm/copy so the work each request does now scales with the size of the request rather than being repeated in full for every item in it. Moving or copying a single item is quicker, and the gap widens with the size of the selection: a bulk move of a dozen notes or files now costs a small fraction of what it did, and a copy of the same selection less again. The same improvement reaches the app wherever you drag, move or duplicate items across Files, Notes and Projects, and it holds for objects shared with a large organization, where the cost previously grew with the number of people who could see them. Behaviour is unchanged in every other respect: the same objects move, the same permissions apply, and every view still refreshes to match immediately afterwards.

2026.08.29 Projects > Adding A Task Or Sub-Goal Is Less Cluttered

Every expanded goal or sub-goal in a project's Tasks & Goals view used to show its own "+ Add new task" (and, for goals, "+ Add sub-goal") row underneath it, so a project with several goals read as a dense stack of extra rows. Those links now appear only when you hover over a goal or sub-goal's row, right on the row itself, so the list stays clean until you need to add something.

2026.08.28 Models > Choose A Reasoning Effort Level For A Model Override

When you override an agent's model for a chat, you can now also choose how much it reasons before answering — Low, Medium, High, Extra High or Max — from a new Reasoning control that appears next to the model picker on the chat screen. It shows up once you have picked a specific model, defaults to Medium, and now applies consistently across every model that supports adjustable reasoning effort.

2026.08.28 Agents > Turn On Every Tool When Creating An Agent

"Enable All Tools" already existed as a per-conversation option under Advanced Chat Settings. It is now also available while creating a new agent, right on the Instructions tab, so a new agent can start with access to every available tool instead of only the default set. A tooltip explains the tradeoff: with 200+ tools available in Sapience, turning them all on can slow an agent down early in a conversation, which is why 50 of the most commonly needed tools are already enabled by default.

2026.08.27 Sharing > Shared Projects No Longer Disappear After Login

A project shared with you could appear for a moment right after logging in or a hard refresh, then disappear from view — even though your access to it was never actually affected. Shared projects now stay visible from the moment they load, on both first login and refresh.

2026.08.27 Chat > Conversations Open At Your Latest Message

Opening or refreshing a conversation showed the very beginning of its history instead of your latest message, so you had to scroll all the way down — sometimes turn by turn on a phone — every time. A conversation now opens exactly where you left off, at the latest message, and stays there even while a diagram, chart or image further up is still finishing loading.

2026.08.27 Chat > Diagrams Render Correctly

Asking an agent for a diagram could show the raw diagram source and a parse error instead of the rendered chart. Diagrams generated in chat — flowcharts, sequence diagrams and more — now render correctly, including ones with multi-line labels.

2026.08.27 Chat > Starter Prompts Appear Immediately In New Chats

A brand-new chat — opened from a floating window or from inside a project — loaded blank, with no starter prompt or suggested questions, until you sent a first message, at which point they appeared alongside it. Starter prompts and suggested questions now appear the moment a new chat is ready, with nothing to type first.

2026.08.27 Files > Project Search Stays Within The Project

Searching a project's Files & Notes tab could return notes from your other projects alongside the ones in the project you were actually searching. Search results are now scoped to the project you searched from, including notes nested inside its subfolders.

2026.08.27 Files > Downloads Are Now Reliable App-Wide

Downloading a file, exporting a chat to PDF, or ending a voice session could occasionally report success without the file actually reaching your device. Every download, export and voice-recording save in Sapience now goes through one hardened path that only reports success once a file has genuinely saved, and a quick dictation into a prompt no longer saves an audio recording to disk unless you ask it to.

2026.08.27 Projects > Refresh From Cloud Is Easier To Find

"Refresh from Cloud" is now the first item in a project's Actions menu, since it is available to everyone with access to the project regardless of edit permission. A matching refresh button also now sits in the project's own header, next to the Chat button, so refreshing a project you are already viewing no longer means opening the menu at all.

2026.08.27 Agents > More Tools Available By Default, And Reported Accurately

An agent's coding and image-generation tools could show as unavailable in the interface even when they were actually usable. Every agent now reports its available tools accurately, including whether it can work in a sandboxed environment. Agents can also now edit notes, list your projects, list a project's files, make web requests, and generate images by default, without needing to be individually configured for it — you can still restrict an agent to a narrower set of tools if you want to.

2026.08.28 API > The Reasoning Effort You Ask For Is Applied On Every Model That Can Take It

Hardened how reasoning_effort on gen3_override is applied across the model range, so a level you set on a request now reaches every model whose reasoning effort can be steered — the Claude, Gemini, Grok, Kimi, Qwen, DeepSeek, GLM and Muse Spark families alongside the OpenAI ones — each in the format that model expects. Where a model has no reasoning-effort control at all, a request now resolves to a stated medium rather than leaving the level unspecified, and nothing is sent to that model which it does not accept. Tier handling is unchanged: a level above what a model supports still lands on that model's ceiling rather than being refused, so xhigh becomes high where high is the top, and max becomes xhigh on OpenAI models. One behaviour change worth knowing if you send gen3_override: leaving reasoning_effort out now means "use the level this agent is configured for" rather than quietly settling on low, so an agent set up to think harder keeps doing so when you override only the model. Effort and temperature are also now decided by a single rule for every model, so a model that accepts both continues to receive both.

2026.08.27 Notes > Attach Files To A Note

A note can now hold files — PDFs, audio, video, or anything else — the same way a folder or a project can. Drag a file onto a note, use the upload dialog, or paste it from your clipboard, and it attaches to the note as its target. Once attached, a file behaves exactly like any other file in Sapience: it opens in the viewer, downloads, is searchable, and keeps its version history. The note body no longer has to carry media inline to hold onto it.

2026.08.27 Sharing > Sharing A Note Or Folder Shares Its Files

Share a note, a folder or a project with someone — or publish a note's public link — and they now see and can open everything inside it at the level you granted, including files attached after the share was made. Folder sharing is supported for the first time, and write access on a shared note, folder or project carries through to write access on the files inside it.

2026.08.27 API > Files API Accepts A Note As A Container

container_uid on the files list endpoint, and parent_uid on all three upload paths, now accept a note's uid alongside a folder or a project. POST /v2/srm/tree anchored at a note returns its files with parent_uid set to the note, and every file anywhere in the tree now reports the container it actually lives in rather than an empty value at the project root. GET /api/v2/files/shared/{token} accepts an optional file_uid so a note's public link can serve its attachments, and downloading one does not count against the note's own view limit. PATCH /api/v2/files/{uid}/metadata now honors write access granted through a shared note, folder or project, not only file ownership.

2026.08.27 Outbound > Outbound Ops Console

Managing outbound campaigns previously depended on a single shared account, so only one person could see how campaigns, messages, people and companies were actually doing at any given moment. A new read-only Outbound Ops Console gives each team member their own sign-in and a live view across all four: campaign health alongside the sending provider's own explanation whenever a campaign is blocked, and searchable, filterable lists of messages, people and companies where clicking a summary number takes you straight to the rows behind it. Every view exports as CSV, Markdown or JSON from the same underlying query, so the numbers a person reads on screen match exactly what a script or report pulls.

2026.08.26 Projects > Templates Shared Down From A Parent Organization Can Be Used To Start A Project

Hardened how a project template published by a parent organization resolves for the organizations beneath it, so a template shared down to your organization can be used to start a new project. Choosing it in the New Project wizard creates your own copy of the whole template: its structure comes across, and so do the files attached to it, which arrive as your copies rather than as links back to someone else's. The new project belongs to you, in your organization, and you can edit it freely from the moment it is created. The template itself is untouched by this — starting a project from a shared template never changes the template other people are still using.

2026.08.26 Notes > Voice Recordings And Link Embeds Come Through Wherever A Note Is Read

A voice recording you dictated into a note, or a link you embedded in one — a spreadsheet, a video, a dashboard — came through as a player or a live embed only while the note was open in the editor. Both now render wherever the note is read: in the note viewer, in a note preview, and on a link you have published. A recording plays in place with ordinary player controls, and an embed loads as itself, on the same published pages that recently gained the app's typography. Neither needs anything extra to load — the player and the embed are part of the note as it is stored, so the note you wrote and the note someone opens from your link are the same note.

2026.08.26 Sharing > Files Now Tell You Who Can Reach Them Too

When listing objects started returning share_data two days ago, files were the one thing left out — a folder or project read told you everything about the notes and tasks it held and nothing about the files sitting beside them, so a client still had to fetch the sharing endpoints separately and join the results itself just for files. That gap is closed. POST /v2/srm/tree and POST /v2/srm/get now return share_data on every file they return, in exactly the same shape they already return it for every other kind of object, so one code path renders both. A file can be reachable in two different ways, and the response now tells you which: a grant made directly on the file, and access that comes from the project the file lives in. Each entry carries grant_source saying which kind it is. A direct grant looks like any other — it has an id, and you can revoke it through the sharing endpoints. Access that comes from the project does not: it has no id of its own, and you change it by changing who is on the project or by moving the file, not by revoking a grant. Read grant_source before you offer someone a revoke button, and before you attribute a grant to a person, because the project-derived kind names the project rather than an individual. Everything else about the field behaves as it already does elsewhere: what comes back depends on who is asking, link tokens are left out for anyone who is not the owner, and an empty list still means nothing is shared while an absent value still means the caller did not ask. Two small additions to the entry shape come with this: the id field can now be absent, and grant_source is new. Both are additive, and a client reading only the fields it already knows about sees no change. Reading a page still costs the same handful of lookups whether the page holds one file or several thousand.

2026.08.24 Sharing > Listing Objects Now Tells You Who Can Reach Them

Reading a single object has told you every live share on it since yesterday, but the endpoints that return a LIST of objects did not. A client showing a folder, a project tree or a batch of notes had to fetch the list, fetch the sharing endpoints separately, and join the two together itself to work out which items were shared and with whom. That join is now unnecessary. POST /v2/srm/get, POST /v2/srm/tree and POST /v2/sapience/tree each return the same share_data on every Sapience object they return that the single-object read already returned: every live grant, covering named-user grants, public links, and access inherited from a tree share on a parent project, with each entry saying whether it is direct or inherited and naming the ancestor it came from. What comes back still depends on who is asking, and link tokens are still left out for anyone who is not the owner. The field keeps the same three-state meaning everywhere: an empty list means nothing is shared, an absent value means the caller did not ask for sharing information, and those remain distinguishable without a second call. Sharing information is also now kept in step with changes across every surface: creating a link, revoking a grant, re-sharing to someone whose access was previously removed, or removing a file from an object all take effect on the next read rather than after a delay. Files are not covered by this and continue to return no sharing information on these endpoints. Reading a page of objects costs the same two lookups whether the page holds one object or several thousand.

2026.08.24 Notes > Published Status Stays Accurate On Notes

The Published toggle on a note could show the wrong state after the notes list refreshed, so opening a note sometimes told you it was not published when it was. Publishing state is now carried through every refresh, and the toggle reflects the note's real state whenever you open it.

2026.08.24 Notes > Restoring An Earlier Note Version Keeps Its Properties

Hardened restoring a note to an earlier version from its version history, so the restored note keeps its full set of properties — title, tags, scope, type and location — alongside the restored content. Restoring a version now changes only what that version actually changed.

2026.08.23 API > One Read Tells You Everyone Who Can Reach An Object

Reading an object told you nothing about who could reach it. The two sharing fields on the object were only ever what a client wrote at save time and were never kept in step with actual sharing, so anything that needed the real answer fetched the object, fetched the sharing endpoints separately, and joined the two by UID itself. GET /api/v2/sapience/{uid} now returns share_data alongside the object: every live grant on it, covering named-user grants, a public link, and access inherited from a tree share on a parent project. Each entry says whether it is direct or inherited, and an inherited one names the ancestor it comes from and how many levels up that sits, so you can show where access originates rather than only that it exists. What comes back depends on who is asking — the owner of the object sees every grant on it, while someone who has been given access sees only their own, and link tokens are left out of the response entirely rather than returned empty. The field also distinguishes "nothing is shared" from "this caller did not ask for sharing information": an empty list means the first and an absent value means the second, so a client can tell them apart without a second call. The older is_shared and share_settings fields are now marked deprecated and continue to return exactly what they always did. Reading an object with fifty grants on it takes the same number of queries as reading one with none.

2026.08.23 Notes > Opening A Note Is Instant, And What You See Stays Current

Opening a note waited on whichever stored copy the app reached for first, so a note you had already loaded could still pause before it appeared, and a note you had changed on another device could show you the older text until you reloaded the page. A note now paints immediately from the copy already on your device, while Sapience checks the stored note in the background and updates what is on screen only if it has genuinely changed since. You get the speed of a local copy and the accuracy of the stored one from the same open, without choosing between them. The check is careful about interrupting you: if you have the editor open it is left alone and your unsaved work is never replaced, and a check that finishes after you have moved on to a different note is discarded rather than applied to the note now in front of you.

2026.08.23 Notes > Notes Shared With You Appear Wherever Notes Are Listed

Places that ask you to pick a note — attaching one to a project, linking one from another note, choosing one from the file browser — built their list from your own notes alone, so a note someone had shared with you could be absent from the list even while it sat in your sidebar. Note lookup across Sapience now runs through a single path that covers everything you have access to, so the notes offered to you are the same set wherever you are asked to choose one, whether you wrote them or someone shared them with you.

2026.08.23 Notes > Renaming Or Re-Scoping A Note Is More Robust

Renaming a note, or changing whether it is private, shared with your organization or global, now always works from the note's current stored copy rather than one the app was already holding. That makes both actions steadier on a note you have edited recently, or one being edited somewhere else at the same time.

2026.08.23 API > Fetch Any Set Of Resources By UID In One Call

Reading something when you already held its UID meant knowing what kind of thing it was first, and calling a different endpoint depending on the answer — one for a note, task, goal, folder or project, another for a file. POST /v2/srm/get takes a list of UIDs and returns a list of resources, whatever mix of kinds those UIDs happen to be, so a client holding an identifier no longer has to work out its type before it can read it. A single fetch is just a one-element list; the request and response shapes are identical for one UID and for thousands, so there is no separate singular call to reach for. Each returned resource arrives in the same envelope POST /v2/srm/tree already uses, which means anything you have written to render tree results renders these too. Anything not returned comes back in a per-UID misses list saying why — not found, deleted, or not permitted — using the same reason codes the individual read endpoints give you, so you can tell a stale link apart from a permissions question without a second round trip. Deleted resources can be included by asking for them. The endpoint returns exactly what the tree endpoint returns for the same UIDs, so the two always agree on what you can see. It reaches the TypeScript SDK as client.api.v2.srm.get().

2026.08.23 Notes > Your Dashboard Activity Feed Shows Your Latest Notes Always

The Dashboard's Activity Feed could show a note from before your most recent edit, since it read from a locally stored copy. It now always reflects your latest notes, matching the freshness already used for recent activity elsewhere in Sapience, so a note you just changed appears right away. for 99% of users this was a non-issue, but for users actively updating notes on multiple devices concurrently this was annoying.

2026.08.22 Sharing > A Published Note Now Looks Like The Note You Wrote

Opening a note you had published to a link showed the words but not the shape: task lists arrived as plain bullets with their checkboxes and ticked state missing, headings and tables carried none of the styling they have inside Sapience, and nothing on the page said what you were reading or who wrote it. A published note now renders with the same typography as the app — headings, tables, code and lists — and task lists keep their checkboxes and show which items are done. Those checkboxes are deliberately read-only, so someone reading your link cannot tick one and be left thinking it saved back to your note. The page also opens with the note's title and description, who owns it, and when it was created and last updated, on Sapience-branded chrome rather than a bare document. The same content styling now applies to PDF and Word exports, so a note you send as a file and the same note you share as a link look like each other, and like the note you wrote.

2026.08.22 API > Ask The Tree Endpoint For Metadata Only, A Page At A Time

POST /v2/srm/tree accepts three new optional fields. Setting projection to "meta" returns only each object's descriptive fields — name, type, scope, parent, timestamps and sharing — and leaves out the stored content, which is what a list view needs and nothing more. Setting depth to "children" limits a read to the direct children of the anchor rather than the whole subtree. And include_counts adds the goal, task, page and descendant totals to project rows. All three default to exactly what you get today, so an existing call returns byte-for-byte what it always did, and there is no new endpoint to adopt. Because they ride on the tree endpoint, a metadata read inherits its cursor pagination: you get next_cursor and has_more back, and you page until has_more is false instead of passing a limit and assuming it covered everything. Where objects carry large bodies the saving is substantial — reading a tree holding 13.8 MB of content as metadata did roughly ninety times less work than reading it in full. GET /v2/sapience/list-meta, which returned at most a fixed number of records and offered no way to ask for the next page, is now marked deprecated: it still works and still behaves exactly as before, and the Sapience apps have moved off it. The tree endpoint also now returns organization-shared objects consistently with the endpoints it replaces, so a whole-workspace read and a scoped read agree on what you can see.

2026.08.22 Projects > Quick Add A Task Without Opening A Dialog

Capturing a task meant opening the new-task dialog, filling it in and saving it, once per task — slow going when three or four occur to you at once. A Quick Add box now sits directly above your task list, on both All My Tasks and a project's Goals & Tasks tab. Type a task name, press Enter, and it appears in the list straight away while the box empties itself and keeps the cursor, so the next one is a sentence and a keystroke away and a run of five takes seconds. On All My Tasks, where a task could belong to any project, a project picker sits beside the box listing your favorites first and then your most recently active projects, and it keeps your choice for the rest of the session so a run of tasks for the same project costs no extra clicks. Each task is created as To Do at medium priority and owned by you; opening it afterwards gives you the full editor for a due date, an assignee, or anything else you want to set.

2026.08.21 Chat > A Menu Button Appears When You Hover An Agent's Reply

The actions on an agent's answer — Copy, Save to Notes, Save to Project, Share and Export — sat behind a small Save & Share link underneath the message, and nothing on screen suggested there was more there than the link itself. Hovering anywhere over an agent's reply now reveals a menu button in the top right of the message, and pressing it opens those same actions at your pointer. Nothing has moved to make room for it: the Save & Share link is still under the message and still does exactly what it did, and right-clicking anywhere on an agent's reply still opens the same menu too. There is one menu behind all three, so the actions you see are identical however you get to it. On a phone or tablet, where there is nothing to hover, the button is simply always shown on an agent's reply, so those actions are one tap away.

2026.08.19 Desktop > One Upload Choice In Windows Explorer, And It Asks Where

The Sapience submenu in Windows Explorer carried a separate upload action for each destination, so you chose where something went by picking a menu item. It now offers a single Upload to Sapience alongside Chat with File. That one action works the same way on a single file, on several files selected together, and on a whole folder, and it always asks where in Sapience you want them: any folder or project in your tree, or the top level. A folder keeps its structure, so its subfolders are recreated and each file lands where it was, including when you send the folder to a project. Chat with File is unchanged, and still opens a new conversation with everything you selected already attached.

2026.08.18 Desktop > Right-Click A File In Windows Explorer To Chat With It

Right-clicking a file in Windows Explorer offered a single Upload to Sapience action. It now opens a Sapience submenu with three choices. Chat with File uploads the file and takes you straight into a new conversation with it already attached, exactly as if you had dragged it onto a chat, so you can ask about a document without opening Sapience first and finding it. Upload to Project lets you pick the project it belongs to and uploads it there, so it is shared with whoever the project is shared with. Upload to Sapience behaves as it always has. Right-clicking a folder, or the empty space inside one, offers the two upload choices.

2026.08.18 Agents > Try An Agent From The Public Catalogue

The public AI Catalogue tells you what each agent and skill does, but the only way forward from a card was to book a call with a Customer Success Manager. Clicking a card now also offers Try For Free, first in the row: it opens a free trial signup in a new tab and, once you have signed up, puts you straight into a conversation with the agent whose card you clicked, ready to type, rather than leaving you on a dashboard to find it yourself. Skill cards carry the same offer and lead to the same signup. The button tells you before you click that it opens a new tab, and the catalogue keeps a link to that tab behind it, so you can open the trial again if your browser blocked the new window.

2026.08.18 Desktop > Local Search Is Much Lighter On Large Accounts

Sapience keeps a local search index so that searching your files, notes, projects, conversations and messages is instant. On a large account that index had been carrying a full second copy of the text it had already indexed, which made building it far heavier than it needed to be. The index now lives entirely in its own background process and no longer keeps that duplicate, cutting its size by well over half on a large library. What you search, and what you get back, are unchanged: full document and message text is still indexed in full, ranking is identical, and the text shown around a match is still drawn from the real content, however far into a long message the match falls. Index building has also been hardened — a slow rebuild now recovers on its own rather than staying in a reduced mode for the rest of the session, searching keeps working even if that background process is unavailable, and the index is released when you sign out. The first time you open Sapience after this update your search index is rebuilt once, after which it is reused between sessions as before.

2026.08.17 Skills > Ask An Agent To Turn What It Just Did Into A Skill

Watching an agent work through something useful and wanting it again next time meant writing the skill yourself. Now you can simply ask: say "turn that into a skill" and the agent writes up what it actually did — the steps in order, the tools it used, the output format, and the traps it hit — and saves it as a skill you own, active straight away and invokable by name. Every agent can do this, so whichever one did the work is the one that captures it, and what you get is a reusable method rather than a summary of the chat. The skill belongs to you rather than your organization, and it appears in your skills list the moment it exists, without a reload. If the name you ask for is already in use the agent says so and suggests one that is free, so a new skill is never left shadowed by an existing one. The agent can also read the new skill straight back in the same conversation, so you can check what it wrote before you rely on it.

2026.08.17 Desktop > Scheduled File Sync Is Now Opt-In, And Runs On Its Own

Automatic file sync in the desktop app is now something you switch on rather than something you switch off. You will find it under Settings, Desktop, Automatic Sync, where you can enable it and pick a 15, 30 or 60 minute interval. If you already had it enabled it stays enabled, on the interval you chose. Alongside that, scheduled syncs have been hardened to run entirely in their own background process rather than inside the app window, so a sync working through a large library no longer competes with what you are doing, and progress reporting is rate-limited to keep the interface responsive while one runs. A sync already under way will not be started a second time, whether by the schedule or by the Sync File System button, and scheduled runs keep their own diagnostic log separate from the app's.

2026.08.17 Desktop > Dialog Close Buttons Sit Where They Should

The close button on dialogs in the desktop app rendered inline beside the title instead of at the top right of the header, while the same dialogs were correct in the browser. Desktop and browser now share one set of styles, so dialog headers, and the rest of the interface alongside them, look the same wherever you use Sapience.

2026.08.17 Files > Your File List Comes Back From Cache Between Sessions

Sapience keeps a local copy of your file list so it does not have to fetch your whole library every time you open the app. On large accounts that copy was not being saved, so each start re-read the full list before the Files page settled. Saving has been hardened to handle the full range of file records, so the cache is written reliably and returning to Sapience shows your files sooner. The larger your library, the more noticeable the difference.

2026.08.15 Notes > Add A Tag From The Ones You Already Use

Adding a tag meant typing it from memory, on a note and again on a project, with no sight of the tags you had already created — which is how one idea ends up spread across roadmap, road-map and roadmap-2026. Clicking Add tag now opens a picker that still lets you type a brand-new tag exactly as before, and lists your existing tags beneath it, most-used first with a count beside each, so the tags you actually use are the ones you see first. Tags already on the note or project are left out of the list, and picking one applies it in a single click. Your tag vocabulary is ready as soon as the app is, and keeps up as you go: a tag you coin on a note is offered to you on a project without a reload. Notes and Projects share the same picker, so tagging works the same way in both.

2026.08.15 Notes > Your Notes Sync Without Re-Downloading The Whole Library

Signing in on a device you had used before re-downloaded your entire notes library every time — on a large account, tens of megabytes of notes fetched in the background before you had opened a single one. The background refresh now asks only for what has changed since your last sync, so a returning user moves kilobytes instead, and the app settles faster and sits lighter in memory, most noticeably on a phone. Everything you had is still there: a note you edited on another device arrives with its changes, a note you deleted elsewhere disappears here too, and notes reaching you through a project someone shared with you keep arriving in the ordinary background sync. Signing in on a brand-new device, or after clearing site data, still fetches your complete library.

2026.08.15 Notes > A Note You Own And Have Shared Appears Once In The Sidebar

A note that was both yours and shared with someone else was listed twice in the sidebar, once for each. It now appears a single time, carrying the sharing details, so the actions offered on the note match what it actually is.

2026.08.13 Models > Grok 4.6

xAI's Grok 4.6 is available everywhere you choose a model: the model selector in chat, the create-agent wizard, the advanced options on a conversation, and the Sapience AI Store. It succeeds Grok 4.5 as the fast generalist of the xAI line, ahead of it on reasoning, coding and agentic work, and keeps the same 500,000-token context window, image understanding and temperature control. The Sapience Agent (Grok) now answers on it. Grok 4.5 stays available and selectable, so an agent already set to it keeps working exactly as before.

2026.08.12 Chat > Sapience loads dramatically faster, especially on mobile

Sapience now downloads only the code a screen actually needs. Opening a chat previously pulled in the diagram renderer, the spreadsheet exporter and the data-grid engine whether or not you used any of them; those now load the moment you first need them and not before. The result is a much lighter app: the sign-in screen loads less than half the code it used to, and the chat screen roughly a third. Phones and tablets benefit most, where memory is tightest and pages that grew too heavy could previously reload themselves. Diagrams, Excel export and all grids behave exactly as before.

2026.08.12 Files > PDFs Preview As The Real Document

Previewing a PDF built a web-page approximation of it first, which stood in for the document's pagination, fonts and vector graphics rather than reproducing them. Opening a PDF now renders the file itself, page by page, so the preview matches the document you would get if you downloaded it. Pages scroll continuously with a Page X of N readout, and zoom in, zoom out and fit-width controls are there for small print. There is no conversion step to wait on, so previews open straight away. Selecting text inside the preview is not part of this first version. Word, Excel and PowerPoint previews are unchanged, and a PDF that cannot be rendered directly falls back to the previous view on its own, so you always get a preview.

2026.08.12 Skills > Skills You're Still Building Stay Private Until You Publish Them

A skill shared with your organization becomes visible to your teammates once its status is Active. While it's still in Draft or Disabled, only you can see it, everywhere skills appear across Sapience, including the Sapience Skills manager.

2026.08.12 Chat > Shrink A Full-Screen Conversation Into A Floating Window

Reading a conversation in the full chat view gave no way to shrink it back down while you kept working elsewhere, so you had to either stay on it or start a new one. A Floating Chat button in the chat header now turns the conversation you're reading into a small draggable card and takes you to the Dashboard, carrying your conversation exactly as it was, including anything still arriving mid-reply. Available for every kind of conversation — direct chats, project chats, file and folder chats, and group rooms — on desktop.

2026.08.12 Skills > A Skill You Own Stays Editable After You Change Its Scope

Changing a skill's scope from Just Me to Team locked it from further editing, even for the person who created it, with no way back in. A skill you own now stays editable by you no matter its scope, and appears read-only to everyone else exactly as it did before.

2026.08.11 Chat > New Conversations Start Straight Away On Non-OpenAI Agents

Starting a new conversation carried a couple of seconds of setup that only agents running on OpenAI models actually use, and it ran whatever model your agent was on — so most new conversations waited for something they would never touch. That setup now happens only when the agent's model needs it: on conversations started with Claude, Gemini, Perplexity and other non-OpenAI agents, which is roughly three in four, opening a conversation drops from about 2.7 seconds to under half a second. Conversations on OpenAI-based agents behave exactly as before, and an agent switched to an OpenAI model partway through a conversation picks up what it needs on its next turn.

2026.08.09 Accounts > The Sign-In Screen Loads A Fraction Of What It Used To

The sign-in screen was downloading the whole application before it could show you a username box, a password box and a button — around 11 MB of JavaScript. Signing in now loads only what that screen actually needs, so it appears far faster, and the difference is most noticeable on a phone or a slow connection. Nothing about how you sign in changes, and the rest of the app still loads as you move into it.

2026.08.09 Mobile > iPads And Sideways Phones Get The Full Layout

Sapience worked out whether to show you the phone layout using more than just the size of your window, so an iPad — and a phone held sideways — could end up with the cramped phone layout instead of the full one. Layout now follows the width of the window alone: an iPad and a landscape phone get the desktop experience, and a narrow window gets the mobile one. The space reserved along the top for a notch or Dynamic Island is now taken from the device itself rather than assumed, so an ordinary browser tab no longer carries an empty band across the top, while phones that draw edge to edge still clear the camera cutout — including full-screen dialogs, which sit outside the main app frame and previously did not.

2026.08.11 Agents > Auto-Generated Memories Are On By Default

Daily auto-generated memories, introduced on 2 August as an opt-in, are now on by default for every account. Once a day an agent reads the messages you have sent since its last pass and saves the new durable facts it finds — your role, preferences, ongoing projects, working style — as Memory objects in the Agent Memories screen, and every agent you talk to afterwards starts the conversation already knowing them. Nothing changes about the memories themselves: they are marked as agent-created, and you can review, edit or delete them like any other memory. A new account is included from the day it is created, with no setup step, and the setting stays controllable per user for anyone who would rather it did not run. The daily summary email is unchanged and still goes only to accounts that asked for it, so everyone else accumulates memories quietly in the background.

2026.08.11 Agents > Project And Task Lookups Cover Everything You Can See

Ask an agent which projects you have, or what is on your plate today, and it resolves the same view Sapience shows you: projects you own and projects shared with you, answered together as one list rather than one of them at a time. Task lists are scoped to live work, so anything sitting inside an archived or completed project is left out — archiving a finished project genuinely clears it from what an agent reports back, and the room that frees up goes to work that is still open. Each task an agent lists now carries its project identifier alongside the project name, so an agent can move straight from a task to the project holding it without another lookup. Where a list is longer than one answer can carry, the agent is told how many it is showing out of how many exist, rather than treating a partial list as the whole.

2026.08.11 Models > Qwen3.8 Max And Grok 4.5

Two new models are available everywhere you choose one: the model selector in chat, the create-agent wizard, the advanced options on a conversation, and the Sapience AI Store. Qwen3.8 Max is built for agentic work across very long inputs, with a one-million-token context window, image understanding, and a reasoning-effort control for trading depth against speed. Grok 4.5 is the faster generalist of the pair, also reads images, and exposes a temperature control for how tightly it sticks to the obvious answer. Either can be picked for a single reply, set as the model behind an agent you create, or browsed alongside the rest of the roster in the AI Store.

2026.08.08 Agents > New In The AI Store: CPTO Agent

The Sapience AI Store adds a Chief Product & Technology Officer agent in the Product & Engineering category. It acts as a full product-and-engineering department for founders without a CPTO, and as a working sidekick for a sitting one: product discovery and roadmap prioritization, PRD authoring using the built-in PRD skill, build-vs-buy analysis, engineering metrics, and architecture strategy. It answers in one voice while quietly handing build work, data analysis, design, research, and contract questions to the right specialist agents behind the scenes, and it opens every engagement by establishing your company context before advising.

2026.08.08 Agents > New In The AI Store: General Counsel Agent

A General Counsel agent joins the AI Store in the Compliance & Legal category. It covers the GC seat end to end — corporate governance, commercial contracting strategy, privacy programs across European, Californian and Australian privacy law, employment-law fundamentals, IP strategy, and legal operations — while always routing tax-law questions to the dedicated Australian and US tax agents and contract redlines to the Contract Review agent, then synthesizing one answer in one voice. Every substantive answer carries a clear professional-advice boundary: it is a powerful working resource, not your attorney of record.

2026.08.08 AI Models > Meta Muse Spark 1.2

Meta released Muse Spark 1.2 quickly after 1.1. It remains an excellent model (fast; large context; multi-modal; great at agentic tool use and delegation). 1.2 gets a small bump in intelligence, material speed improvements and noticeably less verbose reasoning summaries on the wire. One of the best models on the market right now.

2026.08.08 Agents > A New Default Web Research Agent, And Perplexity Gets Its Own

The Sapience Web Research Agent searches with Google from inside Gemini and answers fast, with clickable sources, and can save what it finds straight into your Sapience files as a report — it is the one to reach for on current events, prices, companies and products. The agent previously called Web Research Agent is now the Perplexity Agent, and keeps doing what it has always been best at: deep, multi-source, report-style investigation. Both are in the AI Store under Research.

2026.08.08 Chat > Exported Conversations Say Who Was Speaking

Exporting a conversation to PDF or Word headed every turn either User or Assistant, which reads well enough for a one-to-one chat but not for a conversation with several people and several agents in it, where there was no way to tell who had said what. Each turn now carries the speaker's own name: a person appears as their full name with their username after it, and an agent appears by the agent's own name, so a transcript of a group conversation reads as a conversation rather than an alternating pair of roles. A turn whose author can no longer be looked up still appears in full, labelled by role as before.

2026.08.07 Agents > Sapience Chief Of Staff Agent

A new installable agent puts a chief of staff between you and your workload. It works from your real tasks and projects rather than a guess at them: ask for your daily brief and it reads what is overdue, what is waiting on someone else and what needs a decision from you today — kept to one screen, with each item appearing exactly once — and it saves each brief so the next morning's can honestly report what moved. Send it into a project and it triages what's stuck and writes the missing follow-ups back in, each one flagged for your review rather than silently folded into your plan; say approve and they go live in one word. The things people owe you become chases, assembled as one batch of finished drafts — recipient, subject and body, in send order — for you to release in a single sitting, because it never sends an email or books a meeting on its own, even when told to. Work that belongs to a specialist — finance, marketing, research, data, contracts — is handed to the Sapience agent built for it and comes back as one answer in one voice. A built-in starter schedules the daily brief for 7am every morning. Install it from the Sapience AI Store, in the Agents/Orchestration category.

2026.08.07 Files > Opening A Folder Loads That Folder

Opening a folder on the Files page assembled its contents from whatever the app had already loaded for your whole account rather than asking for the folder itself, so on a large account a folder could show fewer files than it holds — or look empty — until the full listing had finished arriving, with nothing on screen to tell the two apart. Opening a folder now asks for that folder's contents directly and fills them in behind what you are already looking at, so there is no waiting and no spinner. Refresh inside a folder refreshes that folder rather than re-reading your entire library, which on an account of a few thousand files takes about a second instead of several. The top level of the Files page still loads your whole library as it always did, and search still looks across everything you have rather than narrowing to the folder you happen to be standing in.

2026.08.07 Desktop > Your Desktop Session Stays Signed In When Its Sign-In Expires

Leaving the desktop app running long enough for your sign-in to expire signed you out in the middle of a session, and the login screen came back completely empty — even though the app had already renewed your sign-in in the background and had everything it needed to keep you going. A desktop session now renews itself quietly and carries on, so a machine left running overnight is still signed in the next morning. If a sign-in genuinely cannot be renewed — after you change your password on the web, for instance — the login screen comes back with your username already filled in, so there is only your password to type.

2026.08.07 Agents > Faster Web Research, And Several Searches At Once

Ask any agent something that needs current information and it now researches the web through Google Search, which comes back noticeably quicker than before and still attaches the same source links to the answer, so you can see where every claim came from. A question with several strands to it no longer costs one search after another either: an agent can split it into separate searches and run them together in one step, so a three-part question is answered in roughly the time a single part used to take. Nothing changes in how you ask — raise something recent, or say you want current information, and the agent decides to go and look exactly as it did before. Agents that do long-form research reports, which take minutes rather than seconds and produce a written report, are unchanged and still available when you ask for one.

2026.08.07 API > Add, Remove Or Rename A Tag Across Everything That Carries It

Tags could be browsed but not edited, so a tag applied across a hundred notes, projects and files had to be unpicked one object at a time, and near-duplicates like efqm2025 and efqm-2025 could not be merged at all. A single new endpoint, POST /v2/srm/tag, now adds, removes or renames a tag across notes, projects, memories, skills and files in one call: name the objects you mean, or leave them out and it acts on every object you can write that carries the tag. Renaming onto a tag that already exists merges the two into one. Objects you can only read are left untouched and listed back to you in the response, so you can see exactly what was skipped, and dry_run reports the full blast radius before you commit to anything.

2026.08.07 Settings > Advanced Settings Save Automatically

Changing a setting on the Advanced Settings tab meant opening the panel, making your change, and remembering to click Save before you left it — miss that step and the change was lost. Checkboxes and dropdowns there now save themselves as soon as you change them, combining several quick changes into one save and confirming with a brief Saving…/Saved label. This holds however you leave the panel, including pressing Esc or clicking outside it. The Save button stays exactly where it was for Custom Instructions, About Me and your Agent settings, which are unchanged.

2026.08.07 Meetings > Meeting Transcripts Open Ready To Read

Opening a meeting transcript from the Meetings list put you straight into the note editor rather than showing you the note, which had two consequences: links in the transcript's Recordings section could not be followed, because the editor treats a click on a link as selecting it rather than opening it, and a transcript shared with your whole organization showed nothing at all when you clicked it. Meeting transcripts now open in the same note viewer the rest of Sapience uses, so recording links open in one click and organization-level transcripts read normally. Editing is still one click away on the EDIT NOTE button in the viewer; it is simply no longer where you land.

2026.08.07 Voice > The Listening Message Clears When You Stop Talking

Starting a voice conversation posted an "I'm listening" line into the chat, and that line stayed exactly as it was after you finished — still saying it was listening, still showing the animated dots that mean a reply is on its way. Stopping, closing or cancelling a voice session now updates that line in place to say the session has ended and stops the animation, so the conversation reflects what is actually happening. Dictation, which posts nothing into the chat, is unchanged.

2026.08.07 Mobile > Meeting Transcripts Fit The Screen In Portrait

A meeting transcript containing a wide table ran off the right edge of a phone held in portrait, so part of every row was unreachable; turning the phone sideways worked, which made it look like a transcript you could only read in landscape. Tables in a transcript now size themselves to the screen and wrap their text instead of holding a fixed width, so a transcript reads top to bottom in portrait. This applies at render time, so transcripts you already have are covered — nothing needs to be re-saved.

2026.08.07 Notes > A Note You Own Stays Editable After You Share It With Your Organization

Taking one of your own notes and sharing it with your whole organization left you unable to edit it: the Edit button disappeared from the note and the Edit entry disappeared from its Actions menu, with nothing to say why. Your own notes now stay editable by you after you share them organization-wide, and appear read-only to everyone else exactly as they did before.

2026.08.07 Projects > Sorting Goals And Tasks Keeps Each Task Under Its Goal

Sorting the Goals & Tasks list by a column could separate tasks from the goal they belong to, gathering them at the bottom of the list rather than keeping each one beneath its parent. Sorting now reorders goals at the top level and reorders each goal's own tasks underneath it, so the shape of your plan survives the sort whichever column you pick. Sorting the list while it was still loading could also leave rows out of the list until you navigated away and back; sorting now waits for the list to finish arriving.

2026.08.06 Meetings > Your Notetaker Turns Up To The Meetings It Is Booked For

A notetaker booked for a meeting on your connected calendar did not always attend it, and there was nothing on the meeting beforehand to tell you it would not. Sapience now re-checks each booking ahead of the meeting and books the notetaker again if the original booking is no longer live, so a meeting you expected to be captured is captured. Nothing changes in how you set this up: connect your calendar once, and meetings with a join link get a notetaker as they always did.

2026.08.05 Agents > Sapience CMO Agent

A new installable agent covers the whole marketing function: an expert CMO for a company that doesn't have one, and a chief of staff for the one who does. Ask it how to reach a revenue number and it builds the model out loud first — the deals that target implies, the opportunities behind them, how much open opportunity has to be in play to produce them — and only then talks about channels, so you can see whether the plan's own arithmetic holds before you spend against it. Before it answers it establishes your stage, what you sell and to whom, your average contract value, sales motion, current channels, budget and team, then shows you what it assumed and what would change if an assumption is wrong. It helps with demand generation and revenue planning, CAC and payback, positioning and messaging, attribution and what it can honestly prove, and the marketing section of a board pack. Benchmarks arrive with their source and vintage attached, and it will tell you plainly when a figure is a planning aid rather than a measured one. Specialist work — content and SEO, creative, LinkedIn, research, analytics, enrichment, landing pages, agency contracts, ad-claims questions, budget finance — is handed to the Sapience agent built for it and folded back into one answer. Install it from the Sapience AI Store, in the Sales & Marketing category.

2026.08.05 Agents > Content & SEO Strategist

A new installable agent for organic growth: content strategy, technical SEO, and being cited by AI answer engines. Its defining habit is telling you what kind of claim you are getting — whether something is documented by the search engine itself, measured by a named study, genuinely contested among credible practitioners, or simply folklore that gets repeated without evidence behind it. That last distinction matters most in AI search, where a great deal is being sold as settled practice on very little evidence. It establishes your site, CMS, market and current organic position before advising, since the same question has a different answer for a new site than an established one, and it covers content plans and topical authority, the editorial operation, crawling and indexing, site architecture, Core Web Vitals, structured data, international and hreflang, replatform migrations, and what organic attribution can and cannot prove. The CMO Agent hands organic work to it. Install it from the Sapience AI Store, in the Sales & Marketing category.

2026.08.05 Chat > Your Whole Conversation History Loads By Itself

Opening Sapience shows your most recent conversations in the sidebar straight away and then loads the rest behind the scenes. On an account with a long history that second pass could stop short, so the All Conversations list settled at a few hundred entries and only showed everything after you chose Get All Conversations from the Actions menu — and it was back to the shorter list the next time you opened the app. That pass now runs to completion every time, so your full conversation history is there without you having to ask for it. The sidebar still appears immediately, and Get All Conversations is still there if you want to refresh the list yourself.

2026.08.05 Files > Save Now Offers Downloading Or Adding To A Project

Adding a file's copy to a project already worked, but the option lived in a menu that was easy to miss. The Download button in the file viewer is now Save: choose it and pick Download to my computer or Save to a project, and the project picker you already know opens. A file someone has shared with you still downloads in one click as before, since it is not yours to file into a project.

2026.08.05 Files > Share A File With Your Whole Organization From The Share Dialog

Sharing a file with everyone in your organization already worked as Promote to Org File, reached from a file's menu. The Share dialog now offers the same thing directly, alongside sharing with one person or creating a public link: choose Share with all users in my Org and it's done. Once a file is already shared with your whole organization, that option no longer appears, since there is nothing left to share.

2026.08.05 Projects > Calendar And Spreadsheet Join All My Tasks

All My Tasks now offers Calendar and Spreadsheet alongside its table, Eisenhower Matrix and Kanban Board. Filter or search in the table and that same selected set carries into every view; a prominent Filters applied banner makes the narrowing obvious, and Clear All Filters restores your workload without changing the view or the built-in All My Tasks scope. Calendar lets you reschedule tasks from several projects in one place. Spreadsheet shows each task with its project and relevant goal hierarchy, lets you add tasks and goals to a writable project, and supports adding a goal with its tasks in one save.

2026.08.05 Chat > A Conversation You Name Yourself Keeps That Name

Sapience names a conversation for you based on what you discuss in it, and refreshes its summary as the conversation grows. If you had renamed a conversation yourself, that automatic naming could later replace your name with a generated one. Rename a conversation now and the name is yours to keep — Sapience will not rename it again, however long the conversation runs or however many times you come back to it. Summaries carry on refreshing exactly as before, and conversations you have never renamed are still named for you automatically. Conversations you renamed before this change are not covered retrospectively; rename one again and it will be.

2026.08.04 Projects > View All Your Tasks As A List, Eisenhower Matrix Or Kanban Board

All My Tasks can now switch between its table, Eisenhower Matrix and Kanban Board. Apply a filter or search first and the same selected tasks carry into whichever view you choose, including tasks from several projects; changing the view never brings filtered-out work back. You can open and edit a task, drag its priority in Eisenhower or its status in Kanban, refresh the view and use it full screen. Your changes are saved to the project each task belongs to and are still there when you return. The table remains the default, and table-only tools such as grouping, export, column choice and bulk actions stay in the table.

2026.08.04 Email > Emailing An Agent With Attachments No Longer Clutters The Conversation

Emailing a Sapience agent with one or more files attached used to add an extra line to the conversation for every attachment, ahead of your actual email — a message with three attachments could show up looking like four separate messages instead of one. The conversation now shows just your email. Every attachment is still added to the conversation's file list and the agent still knows about all of them.

2026.08.04 Chat > Retrying A File Attachment No Longer Lists It Twice

If attaching a file to a conversation failed partway through, trying again could leave the file listed twice once it succeeded. Attaching the same file again now always leaves exactly one listing for it, however many times you try.

2026.08.04 Agents > Sapience CFO Agent

A new installable agent covers the whole finance function: it's a fractional CFO for a company that doesn't have one, and a chief-of-staff sidekick for the CFO who does. Before it answers anything, it establishes your stage, jurisdiction, entity type, fiscal calendar and revenue model, then shows you a short summary of what it assumed and what would change if an assumption is wrong. It helps with budgets and rolling forecasts, unit economics, cash and runway, fundraising and cap tables, covenants and treasury, month-end close, and board and investor reporting. For deal and valuation work — financial models, comps, precedent transactions, accretion/dilution — it hands the work to the Deal & Corporate Finance Analyst and folds the result into one answer. Install it from the Sapience AI Store, in the Finance + M&A category.

2026.08.04 Agents > Deal & Corporate Finance Analyst

A new installable agent for the person who builds the model — the LBO, the variance bridge, the returns decomposition, the 13-week cash flow, the investment-committee memo. It works two seats and asks which one you're in before doing analytical work, since the same question has a different right answer in each: a PE deal seat (LBO builds matched to the stage of the process, integrating a quality-of-earnings report, returns bridges that decompose IRR and MOIC, SaaS diligence metrics, IC memo craft) and a corporate finance seat (variance analysis, rolling reforecasts, 13-week cash flow, covenant headroom, board-pack framing). Install it from the Sapience AI Store, in the Finance + M&A category.

2026.08.04 Menu > Reorder Your Favorites By Dragging Them

Favorites — files, notes, agents, chats and projects you have starred — always sorted by how recently you favorited them, with no way to put the ones you use most at the top. You can now drag a favorite into the order you want, right on the dashboard Favorites board or in the sidebar list, and the new order carries through everywhere favorites appear, including the sidebar's View All Favorites list. In that list, where entries can sit far apart, up and down arrows do the same job as dragging. If you have never reordered anything, your favorites keep sorting by recency exactly as before.

2026.08.04 Chat > An Attachment You Add Shows In Your Own Thread Straight Away

Files and images you added to a conversation reached everyone else in it, but the thread on your own screen did not always say so until you reloaded: an image you pasted appeared for the other people in the conversation and not for you, and a file you dropped onto a chat, or one you dropped elsewhere and chose to chat about, could go in without the conversation announcing it. The conversation now shows the attachment to the person who added it as soon as it lands, whichever way it was added — pasting an image, using the paperclip, dropping a file onto an open chat, or dropping one anywhere else and choosing to chat about it rather than file it. What everyone else in the conversation sees, and what you see when you come back to it later, are unchanged.

2026.08.04 Projects > Picking A Project Works The Same Way Everywhere

Choosing a project — saving an answer out of a chat into one, starting a task from the New menu, or picking one in the sidebar — asked the same question in three different ways, and each list arrived in a different order, none of which put the projects you actually work in near the top. Starting a task also offered projects that were no place to file work: ones you had archived or finished, and templates. Every place you pick a project now shows the same list in the same order — the ones you have starred first, then the rest with the most recently updated at the top — so what you want is where you expect it wherever you are choosing from, and the star beside a row tells you why it is sitting there. You can type to narrow the list as before. Starting a task now offers only projects that are actually running, so an archived project or a template is no longer somewhere a task can land.

2026.08.04 Chat > Files And Images You Share In A Conversation Stay There, For Everyone

Attaching a file to a conversation, or pasting an image straight into the message box, confirmed on your own screen and went no further: other people in the same conversation were never told it had arrived and could not open it, and a pasted image was read by the agent on the turn you pasted it and then not kept. Both now become a permanent part of the conversation. The conversation announces the attachment by name in the thread, everyone in that conversation is told about it as it happens and can open it without it having to be shared with them separately, and it is still there when you come back and reload — including for someone who joins later and scrolls back through what they missed. Pasted images behave exactly as they did for the agent, which still reads them on the turn you paste them, and are now saved into the conversation as well. Access follows the conversation itself, so an attachment reaches the people in that conversation and nobody beyond it.

2026.08.03 Projects > Start A Project In Another Language (Template Translations)

Starting a project from a template always produced an English project, whatever language you work in, so a German or Arabic team began by translating the whole thing themselves. Pick either of the EFQM templates in the New Project wizard and a language selector now appears offering English, Deutsch and العربية. Choose one and the project arrives already in that language: its name and description, every goal and task, and the text of every note. The project's own agent answers you in that language too, so the work carries on the way it started rather than switching back to English at the first question. What you get is an ordinary project in a single language, not a bilingual one with a toggle — you can rename anything, add to it and share it exactly as you would any other project, and typing your own name or description into the wizard keeps what you typed rather than replacing it. Arabic and Hebrew text lays out right to left wherever it appears: in a note, in a task description, in the calendar, and in a view.

2026.08.03 Projects > Changing Your Mind About A Template Updates The Form

The New Project wizard fills in the name and description from the template you pick, but it only ever did that once: choose a second template and the form kept showing the first one's details, and pressing Finish created the project with them. The form now refreshes to match whichever template is selected, so what you see before you press Finish is what you get. Anything you have typed yourself is left alone — changing template replaces only what the wizard filled in for you.

2026.08.03 Connected Apps > Connect Google Slides to Sapience

Google Slides joins the apps you can connect to Sapience, completing the Google set alongside Drive, Sheets and Docs — it sits in the Individual list, so there is no plan upgrade to make and nothing to ask an administrator for. Open Connected Apps in the sidebar, pick Google Slides and press Connect to authorise your account, or do the same from Settings and Integrations. Once it is connected, Google Slides appears in the app picker beneath the chat box: choose it and Sapience sets out what it can do with your presentations and offers three starting points you can click straight away — summarising your most recent deck, finding a presentation about a piece of work, and building a new deck out of the conversation you are already having. From there just ask in plain language and name the presentation you mean, and the agent will find it, read the content off its slides, or draft and extend a deck on your behalf.

2026.08.02 Agents > Daily Auto-Generated Memories

Sapience can now build your Agent Memories for you. Once a day, an agent reads the messages you have sent since its last pass and saves the new durable facts it finds — your role, preferences, ongoing projects, working style — as Memory objects in the Agent Memories screen, exactly as if you had asked an agent to remember them. Every agent you talk to then starts the conversation already knowing them. The feature is opt-in per user, and memories it creates are marked as agent-created so you can review, edit, or delete them like any other memory.

2026.08.02 Projects > Duplicating A Project Is Now Instant, And Always Complete

Duplicating a project — from the right-click menu on a project card, from the Actions menu inside a project, or by saving one as a template — used to rebuild the copy piece by piece from your browser: every file was fetched and sent back up again, and every note, goal and task was recreated one at a time, so a project of any size took a while and the dialog needed a progress bar to report how far it had got. Sapience now makes the copy in a single step and none of your files travel through the browser at all, so a duplicate arrives in seconds however much the project holds. The Clone Project dialog is simpler to match: it asks for a name and nothing else, because a duplicate now always carries everything — notes, files, goals and tasks — rather than offering four tick-boxes for choosing which parts to bring. Starting a new project from a template takes the same path, so it behaves identically.

2026.08.02 Projects > Task Descriptions Are Labelled Wherever You Hover Them

Hovering a task shows a preview of its description, but what you got depended on where you hovered: the Task List printed a cramped one-line preview behind an abbreviated label, and the Calendar view showed the description exactly as it had been typed, so any formatting in it arrived as punctuation around the words. Every one of those previews now opens with the description's own heading — Task Description, Note Description and so on, matching the kind of thing you are hovering — followed by the description itself as formatted text, so bold, lists and paragraphs read the way they do when you open the item. A description longer than the panel scrolls inside it rather than being cut off at the first line, and an item with no description shows no panel at all instead of an empty box following your pointer.

2026.08.02 Settings > Create A Project Ticks Itself Off Straight Away

The Create a Project step on your Onboarding & Setup tile could sit unticked even when you had dozens of projects, because the tile made up its mind about that step while your projects were still being loaded and never looked again. It now waits for your projects to arrive before deciding, so the step reads correctly the first time the dashboard draws and matches what the sidebar is telling you. The step also now carries the same icon Projects uses everywhere else in the app.

2026.08.02 Projects > Add Goals And Tasks Straight Into The Spreadsheet View

A project's Tasks & Goals Manager offers a Spreadsheet view for working through many rows at once, but it only ever listed tasks: the goals those tasks belong to were missing entirely, and the Goal column beside every task was blank, so the one view built for bulk work could neither show nor set the thing that organises the work. The Spreadsheet view now shows your goals as rows in their own right, with each goal's tasks nested underneath it and sub-goals sitting in their proper place; a task that belongs to no goal appears at the top level rather than dropping out of sight. Goals are editable here like anything else, so you can retitle one or set its owner, status, urgency, importance or due date without leaving the sheet, and changing a task's Goal moves it under that goal as you watch. Add Task and Add Goal links now sit below the sheet: each drops in a blank row you fill in across the columns, you can add as many as you like in one sitting, and a single Save writes them all. A goal you add that way is offered in the Goal column straight away, so adding a goal and filing several new tasks under it is one pass rather than a save-and-come-back.

2026.08.02 Projects > Task Descriptions Read Properly In The Hover Preview

Hovering a task in the Task List view shows a short preview of its description, but that preview printed the description exactly as it had been typed, so any formatting in it arrived as punctuation around the words rather than as formatting. Bold text was the obvious case, since agents use it constantly when they write a task for you. Descriptions in the hover preview now read as formatted text, the same way they do when you open the task itself.

2026.08.02 Agents > Your Agents Know Which Company You Work For

Every conversation opened by telling the agent a little about you — your name, your role, and the name of your organization — and stopped there. A name on its own tells an agent almost nothing: it could not say what your company does or point at its website, so anything it needed to know about your employer had to be typed in again each time. Agents now begin every conversation knowing your organization's name, its website and its description. An agent drafting for a client, sizing up a competitor or writing in your company's voice starts from what your company actually is rather than guessing from a name. The details are read fresh each time a conversation starts, so if your organization's website or description changes, the next conversation you open reflects it. This applies to every agent, the Web Research Agent included.

2026.08.02 Projects > A Clearer Projects List and Project Home

The projects list and each project's home page have been redesigned. The list now opens by saying what Projects are for and links to the documentation, and each project card carries a folder icon, the project's status, and its task, file and member counts on one line that stays readable as the window narrows. Inside a project, the name, status, description and documentation link now sit together in a single header card with the Chat and Actions buttons pinned to its top right, and the tabs beneath belong to that same card rather than floating away from it. Project Information, Quick Stats and Tags each carry an icon so you can pick out the one you want at a glance. The AI Summary panel has been removed from a project's overview: it restated the name, status, file, member and note counts and the creation date already shown in the two cards directly above it.

2026.08.02 Settings > Your Setup Checklist Now Covers Six Steps

The Onboarding & Setup tile on your dashboard listed three things to set up. It now covers six, adding the three that make the biggest difference once the basics are in place: putting an agent to work on a schedule so things happen while you are away, uploading files so your agents answer from your own material rather than the open web, and creating a project so a piece of work has its own home with its own files, goals and agents. Each step is still a single click from the place that fixes it, and still marks itself done as soon as you finish it without a reload. The six now sit as compact cards, three across, so the tile takes up far less of your dashboard than the list did, and hovering any card tells you what that step is for and why it is worth doing. Once all six are done the tile still becomes your tip of the day.

2026.08.02 API > Look Up A Username From An Email Address

The API can now resolve an email address to the Sapience username of the account it belongs to, useful when you have someone's email and need to address them by username elsewhere in the API or SDK. If more than one account shares that email, the response tells you to ask the person directly which account and username to use rather than guessing.

2026.08.01 Settings > Your Dashboard Shows What Is Left To Set Up, Then Teaches You Something Daily

Sapience works considerably better once it knows a few things about you — how you want agents to behave, who you are and what you are working on, and which of your everyday apps it can reach — but nothing on the dashboard ever said so, and most people never filled any of it in. A new Onboarding & Setup tile now sits at the top of your dashboard listing those three steps and marking each one done or not, and every incomplete step is a single click from the place that fixes it: the AI Customization settings for your instructions and your About Me, the Connected Apps manager for your apps. Complete a step and the tile updates as soon as you close the panel, without a reload. Once all three are done the tile does not disappear — it becomes a tip of the day, one from a catalogue of thirty, covering things like translating a whole file, dictating in one language and writing in another, putting an agent to work inside a note with @, and letting the Eisenhower Matrix sort your tasks. Everyone sees the same tip on the same day. Like every other dashboard tile you can reorder it, collapse it or hide it from Customize Dashboard, and that choice follows you to your other devices.

2026.08.01 Chat > Rename, Edit or Delete Any Conversation, However Old

Renaming a conversation, editing its details, deleting it, sharing it or asking it for a summary reached only your most recent conversations — roughly the newest hundred by the date you last touched them, and fewer than that in practice, since conversations you had already deleted took up places in that count without appearing anywhere. An older conversation would open and read perfectly well, then report that it could not be found the moment you tried to rename it. All of those actions now reach any conversation in your history, however far back it sits. Which conversations you can change is decided by who owns a conversation and by the level of access a share carries, so a conversation shared with you to read stays readable and one shared with you to edit can be edited.

2026.08.01 Files > Choose Chat Or Upload When You Drag A File In

Dragging a file onto Sapience always opened the upload form, even when it was obvious where the file belonged, and there was no way to say you simply wanted to talk to an agent about it. The app now only asks when it genuinely cannot tell. Drop a file while you are in a chat and it attaches to that conversation with no clicks at all. Drop one anywhere in Files, at the top level or inside a folder, and it uploads straight to where you are standing and shows you the progress. Everywhere else you get a two-button choice: chat about the file, or upload it. Inside a project the upload button names that project and files it there in one click, and choosing to chat opens a new conversation with the file already attached and confirmed in the thread. Dragging in a whole folder still recreates its structure exactly as before.

2026.08.01 Agents > Fewer, Clearer Categories in the AI Catalogue

The AI Catalogue's category list had grown to twenty-six shelves, several of which were the same idea under two names — data analysis next to Data & Analytics, finance next to M&A, coding next to Product and websites, compliance next to Legal — so browsing for a subject meant checking two or three of them to be sure you had seen everything Sapience offers. Related shelves are now merged, taking the list to eighteen: Data & Analytics, Finance + M&A, Compliance & Legal, Product & Engineering, Sales & Marketing, Writing & Communication, Help & Support and Strategy each hold everything that belongs to them, and each count covers the whole shelf rather than part of it. An agent or skill can belong to more than one category, so a tax specialist is filed under Finance + M&A and also appears under Compliance & Legal. Three agents that had additionally been listed under Research — a contract reviewer and two tax specialists — now appear only under the categories they belong to, so Research holds the research specialists and nothing else.

2026.08.01 Organizations > Sign-Up Asks For Your Company Website

Creating an organization as you sign up now asks for your company website alongside the organization name. Sapience fills it in for you from your email address — sign up as you@acme.com and it proposes your company domain — and it stays yours to correct or replace, so you are confirming a sensible default rather than typing it from scratch. Generic mailbox providers are left blank for you to fill in, since a personal mailbox says nothing about the company. What you type is tidied into a consistent form, so your domain on its own, in capitals, or with a trailing slash all come out the same.

2026.08.01 Chat > Web Research Answers Always Show Their Sources

Answers drawing on web research often listed their sources as a label with nothing after it — the title of a page, then an empty space where the link should have been — or gave no sources at all, even though the research had found them. Which sources appeared, and in what shape, was left to the agent to write out by hand, so it varied from one answer to the next. Sapience now takes the sources from the research itself and renders each one as a numbered marker you can click, so every web source behind an answer is present and clickable every time. This holds both when you chat with the Web Research Agent directly and when another agent does web research for you part-way through an answer, and the sources are kept with the message, so they are still there when you come back to the conversation later.

2026.08.01 Agents > Browse the AI Store by Category, With Agents and Skills Together

The AI Store on sapiencecloud.ai listed agents and skills under separate tabs, each with its own set of categories, so finding everything Sapience offers for a subject like Finance meant looking it up twice and putting the two answers together yourself. The store now has a single AI Catalogue: pick a category once and you see that category's agents and skills in one list, every card marked as an agent or a skill, with an All / Agents / Skills filter if you want to narrow it down. Categories that only ever held skills — presentations, visualisation, planning — now sit alongside the rest instead of being hidden behind a tab you might never open, and each category's count covers everything in it rather than half of it. Agents are listed ahead of skills within a category, and AI Models & Tech stays where it was. Choosing a card from the store now offers to put you in touch with a Customer Success Manager, since running an agent or a skill needs a Sapience workspace to run it in.

2026.08.01 Projects > Share a Whole Project With Your Organization

Sharing a note with your organization has been a right-click away for a while, but a project could only be shared a piece at a time, so the thing that held the work together never travelled with it. Right-click a project card in Projects, or open the Actions menu inside a project, and you can now share the whole project with your organization in one step: its notes, tasks and goals go with it, along with every attached file, and you can make it private again just as easily. You can also publish a project as an organization template so colleagues can start new projects from it, and update a published template later to include anything you have added since. A new Shared with Org filter in the projects list shows what your colleagues have published, kept separate from your own projects so your list does not fill up with everyone else's work. Sharing reaches everything inside a project, so a large one can take a moment; the confirmation tells you what it covers before you commit.

2026.08.01 Connected Apps > Connect Slack to Sapience

Slack joins the apps you can connect to Sapience, in the Individual list alongside Gmail, Notion and Google Drive — there is no plan upgrade to make and nothing to ask an administrator for. Open Connected Apps in the sidebar, pick Slack and press Connect to authorise your workspace. Once it is connected, Slack appears in the app picker beneath the chat box: choose it and Sapience sets out what it can do with your workspace and offers three starting points you can click straight away — catching up on what you missed in a channel, finding the conversation where something was decided, and drafting a message to the team. From there just ask in plain language, naming the channel or the person you mean, and the agent works with Slack on your behalf.

2026.08.01 Chat > Copy the Source Behind a Citation

When an agent cites its sources, each claim carries a small numbered marker, and hovering one showed the name of the source it came from. That was all it did — the label was there to read and nothing more, so following a web source meant retyping it or hunting for it in the Sources line underneath. Hovering a citation marker now shows that label as a panel you can click, and clicking it copies the full source link to your clipboard with a brief confirmation in place of the label. The panel stays put as you move the pointer onto it, so it is there to be clicked rather than disappearing as you reach for it, and it keeps clear of the edge of the window instead of overlapping the text beside it. Citations that point to one of your files copy the file name.

2026.07.30 Projects > A Goal Column and One-Click Filters on Task Grids

Task grids in Views & Reports now carry a Goal column, so you can see which goal a task belongs to without opening it. Quick actions sit above the grid to override the view's default filter in one click — "Show Done" being the common one — and grouping got easier to live with: groups start collapsed, with Expand All and Clear Groups to hand, translated column headers, and a wider Name column.

2026.07.24 Chat > The Agent Tells You It Is Still Working

A hard question can leave a gap of ten to twenty seconds before the first word of an answer arrives. The agent now posts a short sequence of progress lines while it works, so you can see it is active and roughly what stage it has reached.

2026.07.24 Agents > Several Tools at Once on Eight More Models

Agents on eight further models — including the engine behind the default project agent — can now call several tools in the same step instead of working through them one at a time. Answers that gather from a few places at once come back noticeably faster.

2026.07.24 Agents > Deeper Thinking by Default for Project, File and Folder Agents

The default agent that every new project, file and folder gets now runs at a higher reasoning effort: answers take a little longer and come back better considered. The same pass reviewed the tool set these agents ship with, so a newly created project agent can browse that project's files from its first turn.

2026.07.24 API > Domain Restriction Applies on Every Web-Search Path

`POST /api/v2/utilities/web-search`, and every agent's built-in web research, now apply a caller's "restrict to these domains" filter consistently across all search modes, including the default one. Agents configured to draw only on authoritative sources — among them Legal, Contract Review and Due Diligence — inherit this directly.

2026.07.24 Projects > Task Edits Save From Everywhere You Can Open a Task

A task can be opened from a Views & Reports drilldown, from chat, from search results and from a note. All of those entry points now run through the same save path, so an edit made from any of them persists, and the Delete button in the dialog works from each.

2026.07.24 Projects > One Goal Dialog Behind Every Entry Point

Creating or editing a goal now goes through a single shared dialog wherever you start from, so the due date and the assignee are handled the same way on every screen and persist with the goal.

2026.07.24 Files > Upload From the Dashboard Straight Into Any Project

The dashboard's upload tile now opens the uploader directly and lets you choose where the file lands — any folder or any project — using the same nested picker Files uses everywhere else.

2026.07.24 Agents > Approval Carries Through to the Work

When an agent asks permission to create tasks or goals on your behalf, that approval now flows straight into the action it was granted for, so the work happens in the same turn you approve it.

2026.07.24 Files > Hardened Video Processing

Hardened video-to-audio and video-to-text conversion, improving how it handles transient failures and broadening the range of inputs it accepts. The same pass hardened public message-share links and the delivery of task-notification email.

2026.07.24 Sharing > Shared Messages Carry Their Sender Details

A shared message now reliably carries the name of the person who sent it, and sharing by public link handles the recipient list more robustly where there is no named recipient to add.

2026.07.24 Files > Files Keep Their Folder, Tags and Sharing

Hardened how a file's stored attributes are read, so its folder, its tags and its sharing details survive a wider range of stored values and stay attached to the file.

2026.07.22 Models > Google Gemini 3.6 Flash

Google's Gemini 3.6 Flash is selectable in the model picker, and it is now the default model behind chat with a file, a folder or a note — the same capability as before at a lower cost per answer.

2026.07.22 Menu > A Favorites Board on Your Dashboard

Your pinned items now appear as tiles on the dashboard, three across, each with a hover summary of what it is and a right-click menu to open it or drop it from favorites. The sidebar favorites list was tightened up to match.

2026.07.22 Chat > The Model Chip Reads "Agent Default" Until You Change It

The chip under the chat box always showed the underlying model name, which made it look like a choice you had made. It now reads "Agent Default", and names a model only once you have picked one yourself.

2026.07.22 Settings > Turn On Developer Tools

Settings → Advanced carries a new "Enable Developer Tools" switch. Turn it on and developer-oriented views become available on that device, including in the desktop app.

2026.07.21 Projects > Alt+T Creates a Task From Anywhere

Alt+T opens the new-task flow wherever you are, and the project picker inside it opens instantly rather than pausing to load your projects first.

2026.07.21 Files > A File You Just Created Appears in Its Project

Tightened how a project's file list takes up newly created files, so a file created inside a project shows in that project's list straight away, without a reload.

2026.07.20 Skills > Six Skill Categories Instead of Sixteen

The skill picker was split across sixteen overlapping categories. It now uses six clear ones, so finding the skill you want takes a look rather than a hunt.

2026.07.20 Models > A Shorter Model Chip

Five older Claude and GPT models came off the quick model chip under the chat box to keep the everyday list short. All of them are still available when you create an agent or open Advanced options.

2026.07.20 Projects > A Tidier Task Dialog

Goal and Assignee now share a row in the task dialog instead of stacking, and the Description field carries placeholder guidance on what belongs in it.

2026.07.19 Models > Moonshot Kimi K3

Moonshot's Kimi K3 is selectable in the chat model picker, the create-agent wizard and agent settings.

2026.07.19 Models > Reasoning Effort Is Wired Through on Muse Spark

The low, medium and high reasoning-effort setting is now carried end to end into Muse Spark agents, so the level you choose shapes how much the model works a question. Response length across the fleet was tuned at the same time, so a short question gets a proportionate answer.

2026.07.19 Agents > An Agent's Abilities Follow the Agent, Not the Vendor

What an agent can do — file upload, deep research and the rest — is now determined by the agent's own configuration rather than inferred from which vendor its model comes from. Moving a project agent between models keeps its capabilities intact.

2026.07.19 Voice > Hardened Voice in the Desktop App

Hardened dictation and read-aloud in the desktop app, aligning their network path with the rest of the application and adding retry behaviour so a brief connection blip no longer ends the attempt.

2026.07.19 Chat > Live Updates Reconnect Across a Session Refresh

The live-update connection behind new messages and file changes now refreshes its credentials and reconnects on its own when a sign-in session rolls over, so the app keeps receiving updates without a reload.

2026.07.19 Agents > The Skills Tab Fills In as Skills Arrive

The Agent Browser's Skills tab now watches for skills to finish loading and fills itself in when they do, so opening it immediately after signing in shows the full list.

2026.07.19 Files > A Second Grid Engine for the Files Page, in Preview

Settings → Advanced carries an opt-in preview, "Use AG based FL3", which renders the Files page on a different grid engine. It looks and behaves the same; the switch is there so you can try it.

2026.07.19 Agents > Reasoning Models Run Without an Effort Setting

Agents on OpenAI reasoning models now fall back to a sensible reasoning effort when none has been set on them, so an agent created without that field runs normally.

2026.07.18 Chat > Long Conversations Compact Themselves

An agent can now compact its own conversation when it reaches a sensible point to do so, rather than carrying every earlier turn forward indefinitely. Long chats stay faster and cost less to continue.

2026.07.18 Chat > A Large Attachment No Longer Weighs Down Every Later Turn

Attaching a very large file used to carry its full contents into every subsequent message in that conversation. The agent is now told what the file is and reads it on demand through a file specialist, with a note in the chat when it does.

2026.07.18 Agents > Ask Any Agent About a Specific Large File

The tool that lets an agent work through one large file in depth was switched on only for project agents. It is on for every agent.

2026.07.18 Models > Project Agents Move to Meta Muse Spark 1.1

The default model behind project agents across the platform is now Meta's Muse Spark 1.1 — the same quality of answer at a lower cost per turn.

2026.07.18 Agents > An Agent Names the Vendor It Actually Runs On

The vendor shown against an agent now comes from the model registry rather than from a name pattern, so an agent on a non-OpenAI model reports the vendor it is really running on.

2026.07.18 Agents > Hardened Background Job Progress

Hardened how background agent jobs publish their progress, so a slow or unavailable real-time connection no longer holds up the work behind it.

2026.07.16 Chat > Very Long Conversations Compact Properly, and Say So

Compaction in a very long conversation now targets well below the context ceiling rather than trimming just enough to carry on, which leaves more room for the turn in front of you. You also get an explicit notice that compacting a very long conversation can cost some accuracy.

2026.07.15 Settings > Portuguese, Hindi and Malay Work End to End

Choosing Portuguese, Hindi or Malay now changes the language the agent answers in, not only the language of the interface.

2026.07.14 Settings > Brazilian Portuguese, and Your Language Choice Persists

Brazilian Portuguese joins the supported languages, making ten. Language preference is now stored and restored reliably across reloads, so the language you pick is the one you come back to.

2026.07.13 Voice > Dictate Anywhere With Alt+D

Press Alt+D anywhere in the web app and dictate straight to your clipboard, without first finding a text box to speak into. The web Dictate Mode shortcut moved from Ctrl+Shift+D to Alt+D at the same time, so it now sits with the app's other Alt shortcuts.

2026.07.13 Voice > Desktop Dictation Runs on the Shared Recorder

Desktop dictation now runs on the same recorder the rest of the app uses, which improves its reliability and its handling of edge cases — including telling you plainly when nothing was captured.

2026.07.13 Files > One Fetch Path Behind the Files Page

The Files page now loads through a single consolidated fetch path rather than several refreshes running against each other. List completeness and consistency are markedly better on large accounts, including for files created while the page is still loading and for searches run before loading finishes.

2026.07.13 Menu > A Keyboard Shortcuts Dialog

Press ? or / to see every keyboard shortcut the app supports. The same pass tidied the mobile AI settings controls, the file-list buttons, the microphone labels, the note metadata layout, and adding tags.

2026.07.13 Models > GPT-5.6 Terra in the Model Chip

GPT-5.6 Terra is selectable from the quick model chip under the chat box, alongside the other GPT-5.6 variants.

2026.07.13 Agents > A Project Agent Hands Web Publishing to the Specialist

Ask a project agent to publish something as a web page and it now hands the job to the design specialist agent, which comes back with a URL and a password rather than markup in the chat.

2026.07.31 Desktop > Upload To Sapience From A Finder Right-Click

Right-clicking a file in Windows Explorer and choosing Upload to Sapience sends it straight to your Sapience files. On a Mac, the matching Finder Quick Action did not. It now behaves the same way on both: select one or more files — or a whole folder, which arrives with its structure intact — right-click, and choose Upload to Sapience under Quick Actions, and everything lands in your Desktop Uploads folder in Sapience. The app opens if it is not already running, tells you as each item is queued and if one could not be sent, and asks you to sign in first rather than quietly doing nothing when you are signed out. If you have moved your Sapience folder somewhere else, uploads follow it there. Your original files stay exactly where they were.

2026.07.31 API > Polling a Long-Running Job Tells You to Retry Instead of Saying It Is Gone

Polling a long-running job, or cancelling one, could answer as though the job did not exist or did not belong to you when the real cause was a brief interruption that had nothing to do with the job itself. A client had no way to tell that apart from a job that had genuinely finished and been cleared, so the sensible response — stop polling — meant abandoning work that was still running. Both now answer with a temporarily-unavailable status when the cause is passing, and keep not-found and not-permitted for what they actually mean, so your client can retry a job that is momentarily out of reach and give up only on one that is truly gone.

2026.07.31 Chat > Drag a File Onto a Chat and It Attaches Straight Away

Dragging a file onto a chat attached it to that conversation, but it went through the same upload dialog as everywhere else in Sapience — a destination to choose that a conversation does not use, and a button to press before anything happened. Once it finished the conversation looked exactly as it had before, so there was nothing to tell you it had worked. Dropping files onto a chat now attaches them to that conversation on its own: no dialog, nothing to click, a brief indicator while they upload, and a line in the conversation naming each file once it is attached and ready for the agent to read. Spreadsheets dropped this way are prepared for Code Interpreter, the same as when you attach them with the paperclip, and anything too large or of a type Sapience does not accept is refused straight away with a note in the conversation rather than after a long upload. Drop a folder onto a chat and its files are attached one by one, since a conversation holds files rather than a folder tree. Dropping anywhere else is unchanged — the upload dialog still opens with your files already staged and the destination already set to wherever you are.

2026.07.29 Agents > A US Tax And Accounting Specialist Joins the AI Store

The AI Store now carries a US tax and accounting specialist alongside the Australian one; install it from the Store to add it to your agents. It answers on federal tax for individuals and businesses, IRS practice and procedure, state and local tax — economic and physical nexus, P.L. 86-272, apportionment, pass-through entity taxes, with deep profiles for ten states and a way of pointing you at the right questions for the rest — and US GAAP, including the ASC 740 bridge between your accounts and your tax return. It works from the authority a practitioner would actually rely on, citing the Internal Revenue Code, Treasury Regulations, revenue rulings and procedures, case law and the accounting standards down to the paragraph, and it is explicit about where each source sits: an IRS publication is treated as helpful guidance rather than something you can lean on in a penalty defence, and a private letter ruling written for someone else is not authority you can rely on. Figures that reset every year are handled by naming the tax year and the document series the number comes from and telling you to check the current one, rather than quoting a number that quietly goes stale. Before it answers it sets out what it actually knows about your situation, which assumptions it is making and what conclusion changes if an assumption is wrong, and the facts it still needs — so an answer never sounds more certain than the facts underneath it.

2026.07.29 Views > Act on Several Tasks at Once

Working through a list of tasks meant opening each one, changing a single field and closing it again, over and over. Every row in a task view now carries a checkbox: tick as many as you like and a bar appears above the list offering Delete, Set status and Change due date, each acting on the whole selection in one go. Status lists the six states in the order work actually moves through them, and a due date can be set on the whole selection or cleared from it. The rows update in place the moment the change lands, so a view holding several hundred tasks does not reload and you keep your place, and anything that could not be changed is named individually with the reason rather than folded into a count. The same bar is in a project's Goals & Tasks tab, where it appears only while everything you have ticked is a task — tick a goal as well and it steps aside.

2026.07.29 Views > Open a View Full Screen

A view's list can now fill the window. Click the expand control in the view header and the list opens full screen with your search text, sorting, column widths and current selection carried across exactly as they were — it is the same list you were already looking at rather than a fresh copy of it, so nothing you had set up is lost on the way in or on the way back out. Click again to return it to the page. The Refresh link in the same header is now an icon with a tooltip, matching the other controls on that row and leaving more room for the view's name.

2026.07.29 Views > All My Tasks > UX Improvements

Sorting a view by Priority did nothing, and opening the filter menu on that column offered nothing you could pick. Priority now sorts by how urgent and important a task actually is, so ordering by it puts the work that matters most at the top, and its filter menu lists the five levels — Very High, High, Medium, Low and Very Low — to choose from. The Project, Owner and Size columns behave the same way now: sorting, filtering, searching and Excel export all work on the value you can see in the cell, so Size orders by actual file size rather than alphabetically. The count beneath the list read "Showing 127 of 639 rows", where the larger number counted everything fetched before the view's own rules were applied and was nothing you could act on; it now reads "Showing 127 rows" and follows along as you narrow the list with the search box or a column filter. Columns a view starts with hidden — Assignee on task views among them — can now be switched on from the column chooser, and a column with no fixed width no longer collapses to nothing on a narrow screen.

2026.07.28 Meetings > A Meeting With No Speech Says So

When a recording captured no speech at all — nobody joined, or everyone stayed on mute — the note that arrived afterwards was blank, and there was nothing in it to tell you why. Sapience now treats a recording with no speech in it as its own outcome: the note is written as usual and says plainly that no speech was captured, so a meeting with nothing to summarise is easy to tell apart from one you still need to chase. Meetings that did capture speech are summarised exactly as before.

2026.07.28 Files > Images In More Formats Are Described

Sapience reads the images in your files and writes a description of each one, so a chart or a diagram inside a report is searchable and available to your agents alongside the text. Only a handful of image formats were covered, and anything outside that set was passed over without a description — a bitmap dropped into a slide deck, or a scanned PDF whose pages are stored as JPEG 2000, came back with nothing. Sapience now converts these formats behind the scenes so they get described like any other image: bitmap and TIFF files you upload directly, and the JPEG 2000 images that scanners and design tools routinely embed in PDFs. An image in a format that genuinely cannot be read is skipped on its own, leaving the rest of the document fully described.

2026.07.28 Chat > Saving a Chat to Notes Keeps the Answer, Not the Reasoning

Reasoning models narrate their thinking while they work, and that narration is meant to be a live view of the model at work rather than part of the answer. Saving a single message to a note, saving a whole conversation to notes, and reopening a conversation you had left running all now keep the answer and leave the reasoning behind, matching what Copy message already did. Nothing changes while an answer is being written: the reasoning still streams in and still folds away behind the Thinking control, so you can read it in the moment without it following you into your notes.

2026.07.28 Chat > Sending a Message Falls Back When the Live Connection Is Not Up

Sapience can carry a long-running turn over a live connection rather than a standard request, which is what lets you leave an agent working and come back to it. If that connection had gone down, a message sent over it could sit waiting for a reply that was never coming. Sapience now confirms the connection is genuinely up before it uses it, and takes the standard path when it is not, so a turn either goes over the live connection or falls back cleanly rather than waiting.

2026.07.28 Skills > A Skill That Sets Its Own Response Mode Now Applies It for the Whole Turn

Some skills set the response mode they want to be answered in, overriding whatever you have picked in the Thinking menu. That override reached part of the turn and not the rest, so a skill asking for a longer, more considered answer could be sent one way and answered another. The mode a turn runs in is now decided once, up front, and every part of the turn uses that same answer.

2026.07.28 Projects > Goal Assignments and Completions Send an Email

Handing someone a goal recorded the assignment but told them nothing, so the person who had been given it had to notice for themselves. Assigning a goal now emails the person it went to, naming the goal, who assigned it, and the project it belongs to, with a link that opens that project. Completing a goal tells the people with a stake in it — whoever created it, whoever it was assigned to, and the project owner — one email each, and never one to the person who just completed it. Reassigning a goal notifies only the new assignee, and the quiet cases stay quiet: assigning a goal to yourself, clearing an assignee, or handing a goal to an agent sends nothing at all.

2026.07.27 Mobile > Install Sapience on Your Phone or Tablet

Using Sapience on a phone meant opening a browser, finding the right tab and typing an address every time. Sapience can now be installed to a phone or tablet home screen as a standalone app — its own icon, a full-screen window with no browser bar, and a place in the app switcher, with nothing to download from an app store. Open Menu > Apps, which now holds three entries — Desktop Apps, iPad & Android Tablet, and Phone — and pick the device you want it on. The panel works out the exact browser and device you are holding and shows only the steps that apply to it: Safari, Chrome, Edge and Firefox on the same iPhone each reach Add to Home Screen by a different route, and an iPad that reports itself as a desktop Mac is recognised as an iPad. On Android, Chromium browsers get a real Install button, with the written steps kept underneath it because a browser will not offer its own prompt a second time. From the Phone panel you can also text yourself a link: pick your country, type your number the way you would write it locally, and opening that message on the phone brings up the same walkthrough — detected for the phone this time, and working before you have signed in there. The installed app is always named Sapience and opens at the sign-in page.

2026.07.27 Connected Apps > The Dashboard Panel Loads in Seconds, and Tells You When It Can't

The Connected Apps panel on your dashboard confirms each app you have linked is still connected, and it worked through them one at a time — so the more apps you had connected, the longer you sat watching the spinner, and with ten of them it could take most of a minute. Every check now runs at once, so the panel fills in roughly as fast as a single check takes no matter how many apps you have. And if the checks cannot be completed, the panel now says your connected apps could not be loaded and keeps the Manage link available so you can open the connection manager, instead of leaving you with a spinner that never resolves.

2026.07.26 Chat > Reasoning Folds Away Until You Want to Read It

Reasoning models narrate their thinking as they work, and that narration arrived as a run of “💭” lines stacked above the reply — on a turn with several tool calls it could bury the answer you were waiting for. Each stretch of reasoning now folds into one small Thinking… control with an underlined (view) link: click it to read the reasoning, click again to fold it away. It stays folded on its own, so the conversation never jumps around while an answer is being written, and it shimmers gently while the model is still thinking so you can tell live reasoning from a stretch that has finished. Tool activity stays visible in place, with one control per stretch of reasoning between the tool lines, so you can still follow what the agent is doing and in what order. As before, reasoning is shown only while the answer is being written — a saved conversation keeps the answer itself.

2026.07.26 Chat > Focus Mode Clears the Screen for Reading

Settling in with a long answer meant two separate clicks in two different places — collapse the sidebar in one corner, hide the agent description in another — and nothing tied the two together. A single Focus Mode link now sits in the chat header beside About: click it and the sidebar drops to its icon rail and the agent description disappears in one action, handing the answer the width of the window. The link then reads Full UI Mode and puts both back. Everything you actually use while reading stays where it was — the conversation bar with Updated, refresh and Actions is untouched, and so are the header buttons. Focus Mode belongs to the conversation you switched it on in, so opening another chat or moving elsewhere in Sapience returns you to the full layout on its own and you are never left wondering where your sidebar went. The separate More and Less link that used to hide the description by itself is replaced by this one control.

2026.07.26 Settings > Arrange Your Dashboard the Way You Work

Your dashboard is made up of six modules — Favorites, Agent Work, Recent Meetings, Recent Projects, My Tasks, and Recent Files & Notes — and until now every account saw the same ones in the same order. You can now hide the modules you don't use and drag the rest into the order you actually work in, either from the Customize Dashboard link at the bottom of the dashboard or from Menu > Settings > Advanced. Each module has a checkbox and a six-dot drag handle, with up and down arrows if you would rather not drag, and hidden modules stay listed so you can turn them back on. Your layout is saved to your account rather than to the browser you happen to be using, so it follows you to every device you sign in from, and modules added to Sapience later still appear in their intended position without disturbing anything you have hidden. This applies to the dashboard on web and desktop; the mobile dashboard keeps its own layout.

2026.07.25

Agents > Supported Models > Anthropic Claude Opus 5

Anthropic's newest and deepest-reasoning model is selectable in the chat model picker, the create-agent wizard, Advanced options, and the AI Store model browser. It carries a 1M-token context window and reads images. Opus 5 runs adaptive thinking by default — it reasons before it answers — so a hard question takes a little longer to start replying and comes back more considered.

Agents > Ask a Note to Put Its Action Items Into a Project

Open a meeting note, ask for its action items to be added to one of your projects, and you used to be told to do it yourself — go to the project and copy each line across by hand. The note now hands the work to that project's own agent, which creates the tasks for you: one per action item, wording preserved, each landing as "needs review" so you can check them before they count. It works the same way from a file or a folder. Because the project's own agent does the writing, the tasks follow that project's rules and are attributed to it, and items are taken only from what is actually in the document — if it has none, the agent tells you so.

Sharing > Share And Task Notification Emails Tell You What Was Shared

Share a file or a project with someone, or assign them a task, and they get an email about it. The subject names the thing and the person rather than reading as a generic line — "Ken shared the project “Q3 Planning” with you" — and the email links straight to the item: a project opens on its own page, a shared file opens its info card. Project-share emails also say how many files came with the project. Sharing a project sends a single email about the project, not one for every file inside it.

Agents > Every Tool Call an Agent Makes Is Recorded

A conversation recorded which files an agent searched, but not the rest of the work behind an answer — the tools it reached for to get there went unrecorded once the turn was over. Every function and connected-app tool call an agent makes is now written to the conversation record as the turn happens: which tool, in what order, which calls the model issued together in a single round-trip rather than one after another, a short truncated preview of the arguments, and the size of what came back. The per-conversation effort count now reflects the real number of calls, so you can see exactly how an answer was reached — including whether the model worked through its tools one at a time or fired several of them in parallel.

Chat > Watch the Model Think While It Works

Ask a reasoning model something hard and there used to be a silent gap — often five to thirty seconds — between sending and the first word of the answer. The model's own reasoning now reaches you as it happens, line by line, each line marked “💭 Thinking:”, so you can read what it is working through instead of watching a still screen. It works across the models that reason, including OpenAI Sol and Terra, Anthropic Opus and Sonnet, and Meta Muse Spark, and on both live chat and the long-running agent runs you come back to later. Thinking is shown only while the answer is being written — the saved conversation keeps the answer itself, not the reasoning that led to it.

Files > Drag Files or a Whole Folder Anywhere in the App to Upload

Uploading used to mean finding the upload button, then browsing, then picking. You can now drag files straight from Windows Explorer or Mac Finder filesystem windows onto the Sapience window — anywhere in the app, whatever page you are on — and the upload dialog opens with your files already staged. Drop a folder and it uploads as a real nested tree, with sub-folders recreated as you have them on disk. The destination follows you: drop while you are inside a folder or a project and that is where it lands, and drop while you are in a chat and the files attach to that conversation instead. Existing drop targets keep their own behaviour, so dragging onto an already-open upload dialog still goes to that dialog.

Agents > Accounting & Tax > Australian Tax & Accounting Agent

A practitioner-grade Australian tax and accounting expert joins the AI Store, opening a new Tax & Accounting category. It covers both halves of the discipline — tax (income tax, CGT, GST, FBT, Division 7A, trusts, superannuation) and Australian-GAAP financial reporting (AASB standards, Tier 1/Tier 2, tax-effect accounting) — for businesses and individuals alike. It establishes your taxpayer context first, works down the real authority hierarchy from legislation to ATO rulings, and verifies current-year rates with live web research rather than trusting memorized figures, since thresholds change every 1 July. It provides research, not tax advice — and says so, pointing you to a registered tax agent when a decision needs one.

Agents > Ask an Agent for Deep Research, and Get a Report You Can Keep

Ask an agent for deep/serious web research and it can now run a deep, multi-source investigation instead of a quick search — reading widely and working the question before it answers. You get the findings in the conversation as usual, and the agent also files the full write-up as a formatted report in your workspace, in a dated deep-research folder, so the long version is there to reopen, keep, and share rather than living only in the chat. Deep research reads much more broadly than an ordinary lookup, so it takes longer — it is meant for the questions that are worth the wait.

2026.07.23

Agents > M&A - Private Equity > Due Diligence Agent

A senior M&A due-diligence practitioner joins the AI Store, opening the new M&A category. Point it at a target or a data room and it runs a structured, multi-workstream diligence process — financial and quality of earnings, legal, tax, commercial, operational, IT and cyber, HR, IP, ESG — surfaces and risk-scores red flags, and turns findings into decision-grade deliverables: DD summary memos, red-flag reports, exception logs, and document request lists. It tailors the whole process to the deal, whether you are a strategic acquirer, a PE buyer, or working a carve-out or cross-border transaction.

Agents > Legal > Contract Review Agent

A contract-review expert joins the Legal category in the AI Store, alongside the Legal Research Assistant. It reviews, redlines, and risk-assesses commercial contracts — MSAs, SaaS and software agreements, NDAs, employment, supply, leases, licensing — clause family by clause family, and produces a decision-grade issues list with High, Medium, and Low severity ratings. Before reviewing, it establishes the governing jurisdiction and which side of the deal you are on, then frames the review around your objectives, risk appetite, and deal-breakers.

2026.07.12

Agents > Supported Models > Meta Muse Spark 1.1

Meta's new agentic model is selectable in the model picker. It speaks Meta's OpenAI-compatible Responses API natively — the same streaming path Sapience already uses — so it rides the existing pipeline rather than a proxy. 1M-token context, served without history compaction.

2026.07.11 Models > OpenAI GPT-5.6 — Luna, Terra & Sol

OpenAI's GPT-5.6 generation lands the day it shipped, across three tiers: Terra (the daily workhorse, now the model behind the front-door Sapience Agent), Sol (frontier reasoning), and Luna (fast and cheap). All three carry a 1.05M-token context window. Sol is reserved for licensed users; everyone else is transparently served Terra.

2026.07.11 Voice > OpenAI Realtime Voice 2.1

The realtime voice model moves to gpt-realtime-2.1, selectable under Voice Settings — smarter, though less chatty than its predecessor. OpenAI retired the 2.0 model id on July 6, so anyone who had opted into the previous version is migrated automatically.

2026.07.11 Connected Apps > Google Docs, Microsoft Excel & Microsoft Teams — and Google Drive Goes Full-Featured

Three new apps you can connect, and Google Drive graduates from a token integration to the full surface — agents can browse, read, and act on Drive content on your behalf.

2026.07.11 Desktop > Quick Chat Opens In Its Own Window

Quick Chat used to take over whatever you had open. It now opens in its own focused window, and defaults to the main Sapience agent instead of whichever agent you last used.

2026.07.10 Models > Anthropic Fable 5 (Enterprise)

Fable 5, currently the strongest model available, appears in the chat model selector as an Enterprise entry.

2026.07.10 Connected Apps > HubSpot

Agents can search, create, and update HubSpot contacts, deals, notes, and tasks on your behalf.

2026.07.10 Files > Images Inside Files and Notes Are Understood

Screenshots pasted into a note and diagrams embedded in a PDF, Word, PowerPoint, or Excel file used to be invisible to the system — search and summaries simply skipped them. They are now automatically described on ingest, so both finally see them.

2026.07.10 Agents > Effort Is Recorded Per Conversation

Conversations now record how many steps and tool calls the agent used to reach its answer, so you can see how hard a given turn actually worked.

2026.07.10 Desktop > Updates Install Themselves

Desktop updates previously meant a three-click installer wizard, which most people skipped — leaving stale versions in the field. Updates now download in the background, install silently, and relaunch.

2026.07.09 Skills > Financial Models

New skill: builds a genuine Excel workbook with DCF, sensitivity, scenario, and Monte Carlo analyses — not a table rendered in chat. Also re-pointed to the Daytona sandbox, which removed the minutes-long hangs it used to suffer.

2026.07.09 Skills > Strategy Advisor, PRD Writer, Editor & Email Drafter

Four new expert skills. Strategy Advisor applies established frameworks (SWOT, Porter's Five Forces, matrices) to a strategic question; PRD Writer interviews you first, then publishes a rigorous, testable Product Requirements Document as a report; Editor refines existing text while preserving your voice; Email Drafter writes targeted messages and can send them straight to your inbox.

2026.07.09 Skills > Enterprise-Only Skills

A skill can now be flagged Enterprise-only. Everyone can still see it and read what it does — using it on a lower plan shows an upgrade prompt rather than hiding the skill entirely.

2026.07.09 Agents > Coding Sandbox On By Default, Plus a Fast Python Tool

The full run-code-like-a-developer sandbox is now enabled for every agent rather than opt-in, and a new lightweight Python tool handles quick jobs (crunching a CSV, say) without paying for the heavyweight sandbox spin-up.

2026.07.09 Connected Apps > No More False "Connection Failed"

Connecting Google Sheets and its siblings flashed a red error even when the connection had succeeded. Fixed.

2026.07.08 Skills > Deep Research & Data Storytelling — Now Global

Two skills promoted to everyone. Deep Research investigates a topic across many independent, credible sources and publishes a synthesized, fully cited report; Data Storytelling turns raw data into a narrative with the key insights up front.

2026.07.08 Skills > Reorganized Skill Picker

The skill list had grown into one long flat scroll. Skills are now grouped by category behind a fly-out menu, with type-to-filter search.

2026.07.08 Skills > Skills Can Set Response Depth For a Turn

A skill can now raise the response depth for the single message it runs on — Deep Research, for example, answers in detailed long-form mode for that turn — without you touching any settings or having the change persist.

2026.07.08 Organizations > Skill Instructions Are Protected

End users can invoke a shared or global skill but can no longer read or copy the underlying prompt that drives it.

2026.07.08 Projects > "Task" and "Meeting" in the Global New Menu

Add a task to any project, or start a meeting, from anywhere in the app rather than only from inside the project.

2026.07.08 Voice > Push-to-Talk in Chat

Hold Alt, speak, release — your words appear in the chat composer.

2026.07.08 Voice > Instant Desktop Dictation

Alt+D used to upload and save a recording before giving you text. It now transcribes on the spot, cutting seconds off every single use.

2026.07.08 Files > File Version on the Info Card

The file info card now shows which version you're looking at.

2026.07.07 Chat > Clickable Citation Pills

Web-research citations rendered as raw reference markers. They are now clickable footnote pills that take you to the source.

2026.07.07 Organizations > Per-Org "No Global Content"

An organization can now hide every global agent and skill and curate its own set instead of inheriting the platform catalogue.

2026.07.07 Notes > Desktop Spell-Check Suggestions

Right-click a squiggle in a note on the desktop app to get correction suggestions.

2026.07.07 Notes > Summarize Works on Every Note

Notes containing pasted screenshots or audio clips could fail to summarize — and in some cases crash the chat. Both fixed.

2026.07.07 Projects > Templates That Actually Share

Global templates are now visible to everyone and org templates to the whole org, and "Save as Template" no longer fails outright on a project that contains files.

2026.07.07 API > SRM Summarize

Summarize-and-persist, for any resource kind, through the polymorphic Sapience Resource Manager API.

2026.07.07 API > SRM Tree

One call now returns your whole resource tree — or any subtree — across files, folders, notes, and projects uniformly. The web app's file views read through it.

2026.07.07 Agents > Agent List Polish

Scope badges show whether an agent is personal, org-wide, or global; the refresh button genuinely refreshes; tooltips render properly.

2026.07.07 Menu > Announcement Banner & Announcement Emails

Sapience can now post an in-app banner and send a platform-wide announcement email — used for these very release notes, and for the Tips series.

2026.07.06 Email > Sapience Email Now Sends From sapiencehq.com

Many enterprise mail filters treat .ai domains with suspicion, so agent and notification email now sends from a .com address for better deliverability.

2026.07.06 Agents > One-Click Chat With Any Project

The New Chat browser gains a "Projects" section that drops you straight into a full-screen chat with that project's agent.

2026.07.06 Sharing > Share Modal Upgrades

An access-level column, a "Revoke Access" action on the Users list, clearer errors, and sharing details that survive a refresh.

2026.07.06 Notes > Resizable Images, Edit Note & a Mobile Card Menu

Images in notes can be resized, an "Edit Note" quick action was added, and note cards finally get a proper menu on mobile.