Agent Knowledge is where you give an agent the content it retrieves from to answer accurately — a mix of reused knowledge sets and sources added directly on the agent.
Overview
The Knowledge table lists everything attached to the agent — both the knowledge sets you've linked and individual knowledge sources (documents, websites, apps, data catalogs) you've added directly. Once attached and indexed, the agent pulls relevant pieces from this knowledge when it answers. You can Add Knowledge Set to reuse an existing knowledge base, or Add Knowledge to create a new source on the agent.
Attaching an Existing Knowledge Set
Use Add Knowledge Set and pick one or more existing knowledge sets to attach to the agent. This is the reuse path — build a knowledge set once and share it across several agents. Attaching links the set to the agent; removing it later just unlinks it, so the set itself isn't deleted.
Adding a Knowledge Source Directly
Use Add Knowledge to create a source on the agent. Pick a type:
Type | Description |
|---|---|
Upload Documents | "Upload .pdfs, .doc(x), .ppt(x), xls(x), code files etc." |
Import a Website | "Sync website's content at a defined frequency." |
Application | "Connect to fetch information from external application." |
Data Catalog | "Connect data catalogs with your agents…" |
The source is then indexed so the agent can retrieve from it. These are the same source types you'd add inside a standalone knowledge set — see Add Knowledge Source.
How Indexing and Retrieval Work
Indexing runs the same pipeline regardless of where you add the source: parse → chunk → embed → store. Once the source shows as synced, the agent's retrieval step can match against it — retrieval rewrites the user's question, searches the vector store scoped to this agent's knowledge, re-ranks the candidates, and passes the top chunks (with citations) to the answer model. For permission-aware application sources, users only retrieve content they can access in the source app. See RAG Pipeline for the full end-to-end mechanics.
What's in the Knowledge Table
Each row shows the source's fields and, for directly added sources, a toggle to include or exclude it:
Column | Description |
|---|---|
Toggle (no column header) | The toggle at the start of each row includes or excludes a directly added source without removing it. Only appears on directly added sources, not on attached knowledge sets. |
Knowledge Name | The source's name, with an icon for its source type. |
Error icon (no column header) | Shown on a source that failed to sync. Open it to see the error details. |
Status | Its indexing state. |
Type | Application, Website, Document, Data Catalog, or Knowledge Set. |
Tags | Labels applied to the source. |
Owner | Who added the source. |
Last Updated | When it last changed; hover for the exact timestamp. |
Two more columns — Created On and Last Updated By — exist but are hidden by default. Turn them on from the table's column-visibility control. The empty state reads Add Knowledge — "Add documents, apps, websites or data catalogs to your knowledge."
Removing Knowledge from an Agent
It depends on what the row is:
A source added directly to the agent (documents, a website, an app, a data catalog) offers Delete, which permanently removes it — the dialog warns this can't be undone.
An attached knowledge set offers Remove, which only unlinks it from this agent — the set stays intact for other agents.
Note: The row toggle likewise only appears on directly added sources, not on attached knowledge sets.