Intelligence

Maintain the Skills, structured data, files, reference material and Memory that give an Agent reusable project context.

Intelligence is the maintained context and reusable capability behind an Agent.

If you are setting up a new Agent, start with Build an Agent, then use these surfaces for the specific material the Agent needs. The Give the Agent context page explains how they fit together.

The current product surface has five areas:

• Memory — persistent project context used across Agent work;

• Skills — reusable Agent operating methods;

• Tables — maintained structured working data;

• Files — project working material and artefacts;

• Knowledge — maintained reference material for retrieval.

These are deliberately separate because they have different maintenance and authority models.

Pick the right home

Use the source that matches the thing you are maintaining:

• exact live business record → source system through a Connection;

• recurring operating method → Skill;

• structured mapping or working dataset → Table;

• project artefact or working material → File;

• maintained reference material → Knowledge;

• persistent contextual observation → Memory.

Do not turn Intelligence into a second ERP by copying authoritative data into several places without a clear owner and refresh model.

Intelligence is part of Agent setup, not a separate knowledge project

A builder should be able to explain why each Intelligence item exists in terms of the Agent's work.

If a Table is never used, a Knowledge Base has unclear authority, a Skill repeats global instructions, or Memory is being used to hold exact operational records, the setup probably needs simplification.

Likewise, adding more context is not always the answer. Give the Agent enough context to find what it needs, then test whether it actually uses the right source on representative tasks.

The visible surface is also a builder inventory

Tables, Files, Skills and Knowledge expose list/search/management views so builders can see what exists in the Project. Memory also exposes its layered context model and onboarding guidance in an empty Project.

Those onboarding actions help seed context. They do not mean every connected source or uploaded document becomes unquestioned truth automatically.