This section covers everything an agent knows and remembers — the documents and data it retrieves from to answer accurately, and the facts it carries forward from past conversations.
Overview
Knowledge is what an agent retrieves from at answer time: files, websites, and connected apps, indexed into searchable chunks through a fetch → parse → chunk → embed → store pipeline, then retrieved, reranked, and cited when a user asks something. It can be attached directly to one agent or built once as a reusable Knowledge Set shared across many. Memory is a separate axis entirely — instead of documents, it's what the agent learns about conversations: facts, events, and procedures extracted from what users say, scoped by Memory Profiles and bundled into reusable Memory Sets.
What's in This Section
Topic | Covers |
|---|---|
Attaching knowledge sets and sources directly to an agent. | |
Building reusable knowledge bases shared across agents. | |
Connecting documents, websites, apps, and data catalogs. | |
Watching indexing progress and fixing stalled sources. | |
How content is parsed and chunked for retrieval. | |
The full retrieval story, end to end — from indexing to citations. | |
Enabling memory and understanding how it's read and written. | |
Bundling memory profiles for reuse across agents. | |
Defining exactly what an agent remembers, for whom, and for how long. |
Knowledge vs. Memory
The quickest way to tell them apart: Knowledge is external content you feed the agent — documents and data it didn't originate. Memory is what the agent learns from the conversations it has itself, stored and recalled without you having to upload anything.