Modes: Deep Research and more – Sapience

💡

The mode you use when you chat with an Agent has a large impact on how it behaves and the output you get. Read below for a description of each mode. Note that when you hover over the modes in the dropdown it will also give you an explanation.

Output Modes

In the main Chat interface, to the right, there is a menu then lets you select the output mode you want. This has a major influence on what you will get back from an Agent. See this screenshot:

Balanced (Default)

A balanced mode, that should give you 200-300 words of content in response to your query, as well as a series of suggestions that you can click on if you want to continue drilling into related topics.

Detailed

This is going to give you 2+ printed pages worth of material in response to your query, and will usually include a summary table at the end. In Detailed mode, the Agent will run through its thinking loop for longer, and you will generally get a more detailed, information-rich report.

Q&A

For rapid-fire, bullet-point based responses. This is well suited for embedding in websites and other places where you want terse, summary-styles answers (without sacrificing quality).

Deep Research

Sapience Agents that support Deep Research are capable of writing long, detailed, information rich reports for you. These reports can run to 20-30 pages. When you select Research mode in the chat interface, the Agent will ask you a series of clarifying questions. Its worth answering these properly, as it will steer the agent in its research run.

Some things to keep in mind:

not all Agents can do Deep Research. Check the Agent Features card.

A Deep Research run can take 15 minutes or more. But its worth the wait. The Agent will often make dozens of search queries to the web as well as to data that the Agent has access to. You can watch it work.

Pro-tip: have a regular conversation with the Agent in Default or QNA mode before you kick off the Research run. This is called “priming” the conversation. You will notice that if you have a conversation before doing Research, the clarifying questions it asks you will be far more on-point than if you run Research immediately in a chat.

Visual

If an Agent is designed to do data analysis, then it will usually have data files baked into the Agent definition and advertise Visual Mode as one of its Tools/Features, and will have a mode option in the dropdown for Visual. In this mode, the model will use the Sapience Coding Agent behind the scenes and is steered to produce at least one visual (usually a static PNG image).

Pro-tip: tell Visual Mode that you want it to create an “interactive dashboard, with at least 3 modules. I want the 1st module to show ABC, the 2nd to show XYZ, and the 3rd to show 123”. But be specific about what you want!