Skip to main content
A knowledge store is a vector-indexed corpus tied to your organization. One agent can attach to one knowledge store; many agents can share the same store.

​
Lifecycle

  1. Create — POST /v1/chat/knowledge-store with a name and description.
  2. Upload — Push PDFs, Markdown, or HTML. Agntix chunks (~1k tokens), embeds (Cohere), and indexes (Qdrant).
  3. Process — Documents are processed asynchronously. Subscribe to the knowledge_store.updated webhook to know when a doc is ready.
  4. Attach — Set config.knowledgeStoreId on an agent.
See the API reference for the full schema.

​
Best practices

Keep stores narrow

One store per topic gives much better retrieval than one giant store with everything.

Refresh on a schedule

Re-upload changed documents weekly; old chunks are replaced atomically.

Enable reranking

Cohere reranking adds ~150ms but cuts hallucinations dramatically.

Cite sources in prompts

Tell the agent in its system prompt to cite chunk titles to build user trust.