Knowledge base
The documents your AI answers from — with citations.
Collections and documents
Knowledge lives in collections of documents: pasted text, uploaded files, or content synced from other features. Documents are chunked and embedded for retrieval.
Grounded answers
When the chat answers from knowledge, it retrieves relevant passages and cites them by number. Coverage is honest: if retrieval finds little or nothing, the answer says so instead of inventing.
Ingestion options
Text and common file formats ingest directly; heading-aware chunking keeps structure. Collections can pin a specific embedding model, and re-indexing runs in the background when settings change.
Managed collections
Some features maintain their own managed collections automatically — published documentation, for example, is kept in lockstep with what is publicly live, so answers never cite unpublished content.