Agents
This is NOT consumer chat. Learn about task-expert Agents here.
1. Agents Overview - What’s an Agent?
Agents are at the heart of Sapience. This is NOT consumer chat. Learn about task-expert Agents here.
What is an Agent?: we will side-step the debate raging in AI circles on this topic and tell you what an Agent is in Sapience-world. An Agent is a combination of: 1) an AI model; 2) a reasoning loop; 3) memory; 4) data; 5) tools/skills.
Expand this for more details:
AI Model: this could be the latest Open AI model like GPT-5.2 or their reasoner o3, Anthropic’s latest workhorse (Sonnet) or their reasoner Opus, or a Google Gemini model. We also use bespoke models for certain tasks (e.g. for computer vision, or voice transcription).
Reasoning loop: simple chat systems give you a “completion” or a response. Agents will work in a loop until some “win” condition is met. When you give a Sapience Agent a task, there might be 5, 10 up to 30 “turns” going on between the Sapience system and the model in question (more in the case of agent teams).
Memory: an AI Agent needs to remember things between conversation turns, between conversations, and “forever”. This is a core feature of Sapience’s 4-tier memory architecture.
Data: mission critical. Most consumer chat like Chat GPT are awful at this. Sapience integrates with more than 1000 data sources, and 3000 file types. Its the combination of data + AI that makes Sapience sing.
Tools/Skills: this is a deep topic, but think of a skill as things like “look up a database”, “send me an email”, “edit this file”. These are generally ways of making an Agent be wired into other things. Under the hood this can be MCP, REST, API calls, shell calls - there is a long list.
2. Global Built-in Agents
All of these Agents come “out of the box” when you start Sapience, or are what we call Global scoped Agents. Remember to hit the feature card in the Agent header of the chat interface to learn the specific skills of any agent.
2.1. Web Research Agent
The Sapience web research agent is your best friend when it comes to doing web research. Under the hood, the Web Research Agent has access to:
- the Google search index
- Open AI’s web search tech;
- Perplexity;
- Google’s Search API
- Claude Agent’s SDK web research;
It will intelligently select the best subsystem for the query, and often run multiple queries behind-the-scenes to help you with your task. This Agent is the default used when you run a web search from Sapience Dashboard.
2.2. Sapience Agent
Your Sapience concierge. This Agent has access to many useful tools, like the ability to search all of your files, browse the Agents available to you and tell you what they do, and it can also call the Sapience Support Agent to help with “How do I…” type queries.
This Agent is the quintessential “jack of all trades, master of none agent”.
One cool thing it can do is interrogate the list of available Agents you have in your workspace, and make recommendations on which one to use.
2.3. Data Analyst Agent
Ever wanted a data scientist at your beck and call 24x7? That’s what the Sapience Data Analyst is. Behind the scenes it uses the Sapience Coding Agent to spin up a container (isolated server environment) to work at your behest. Its great at analyzing tabular data (think Excel files or CSV or JSON). You can ask it to analyze data, create visualizations, or even get it to create interactive visuals.
Pro-tip: ask it to create and interactive dashboard based on some data you’ve given it. BUT, make sure you’re clear on what you want it to do. This Agent works best when you’re clear about the structure of the data, and what output you want. E.g. “Use the data about rental income to create a chart showing revenue over time per property, using a color code with a key on the chart”.
2.4. Sapience Support Agent
Not directly callable, but there in the background (like the Sapience Coding Agent) and can be invoked by asking all the Sapience Agent or other Gen3 Agents for help/support.
2.5. Sapience Coding Agent
Like the Support Agent, this is another Agent that is built into the system but you don’t invoke directly. For any Agents that have Visual Mode, this is what is being used. Also any time you’re analyzing data. You can tell if an Agent has access to this by checking the Agent Features.
2.6 File & Project Agents
Sapience can spin-up an Agent that reads/analyzes a given file right from the File Info card. Think of this as single-file retrieval augmented generation (RAG). Same thing with Project Agents, although these are very powerful (see below).
3. Agent Skills & Features
Tools in Sapience manifest as Agent Skills, or Agent Features. Read this dedicated article to learn more:
4. Organization Scoped Agents
Org scoped Agents are agents that have been created by your org admin, and have been published for all your Org users.
- Org scoped agents can’t be described in the abstract as every single Org is different;
- In the Agent Browser, you will see Agents marked with scope = Org. This means that it was created and published by your Org Admin and its worth looking at it.
- Make sure to read the Agent Welcome Message, and check the Agent Features card in the chat header to learn the details of each Agent.
5. Custom Agents
Get in touch if you need a custom Agent created.
- Need an Agent created? Get in touch either with your Org Admin, or the Sapience Team directly and we can help you.
- Sapience works best when its customized to YOUR workflows and is using YOUR data!
6. Project Agents
One of the things that makes Projects great, is that they’re secretly Agents as well!
Whenever you create a Project, you’re automatically creating a Project Agent under-the-hood. This Agent automatically knows about all the Tasks and Goals in the system, can be given custom instructions, and can access all the Project Files and Project Notes.
Read this separate article on Project Agents to learn more: