Knowledges

Knowledge connects an agent to relevant source material without placing the entire corpus in every prompt.

Explore knowledge

Page	Learn how to
Load documents	Read, normalize, and chunk source files.
Embeddings	Turn documents into searchable vectors.
Vector stores	Build, query, update, and choose an index.
Metadata filters	Enforce tenant, visibility, and content boundaries.
Automatic retrieval	Add relevant documents to every agent turn.
Search tools	Let the model decide when and how to search.

The retrieval flow

Sources → load and chunk → embed → index → filter and search → model context

Ingestion prepares the index outside the request path. At runtime, the current prompt searches that prepared index and sends only relevant documents to the model.

Choose how retrieval runs

Use automatic retrieval when most prompts need supporting knowledge. Use a search tool when retrieval is optional or the model may need to refine its query.

Use static agent context instead when the source set is small, stable, and safe to include in every run. Use tools for live account data, permissions, or actions rather than treating a vector index as an operational database.