Skip to main content
Knowledge is what your Agents read. Documents you upload, sources ingested through Foundry pipelines, and structured facts stored as memory or graphs. Knowledge is separate from memory — knowledge is authored (you curate it); memory is learned (agents accumulate it from experience). Both feed into retrieval.

​
Where knowledge lives

  • Document collections — the primary store. Chunks are indexed for retrieval.
  • Memory spaces — see Memory.
  • Causal graphs — structured entity-relationship data. See Causal Graph.

​
How to get knowledge in

Upload documents

Drag files in, or paste URLs. Handles PDF, DOCX, MD, HTML, CSV.

Foundry pipelines

Automated ingest from cloud drives, databases, and APIs.

​
How to organize it

Tag & organize

Group by department, audience, feature, or freshness.

Versioning & supersede

Retire outdated info without deleting the audit trail.

​
How to use it

Attach to agents

Give an Agent access to specific collections, tagged subsets, or memory spaces.

​
Deletion vs. supersede

You can hard-delete a document, but this is rarely what you want. Better: supersede it. Superseded documents:
  • Are excluded from all retrieval.
  • Stay queryable for audit.
  • Can be reversed if you superseded by mistake.
Reserve hard delete for legal/compliance takedowns and PII removals. See Supersede.

​
Storage & limits

  • Free workspaces: 500 MB of source documents, 10k chunks.
  • Team plan: 10 GB, 500k chunks.
  • Enterprise: unlimited (fair use), scales to hundreds of millions of chunks.
The storage counter in Workspace Settings shows current usage broken down by collection.