Skip to main content
A knowledge base is a set of sources the agent can search while it talks, so it answers from your information instead of guessing.

​
Sources

A website crawl follows links from the start page, stays on the same host and respects robots.txt. You can limit it with max_depth and max_pages.

​
Status

A source is pending, processing, ready or failed. A failed source carries the reason in error. A knowledge base is ready once every source is indexed.

​
Keep it current

Use POST /v1/knowledge-bases/{id}/sources/{source_id}/sync to fetch a website again or re-read a file after it changed.

​
Attach it to an agent

Add the knowledge base id to knowledge_base_ids in an agent version. Each knowledge base must be ready when you publish that version. The agent then searches it with a search_knowledge tool.

​
Test what it finds

GET /v1/knowledge-bases/{id}/search runs a query and returns the matching passages. Use it to check that a question finds the right text.

​
Tips

  • Put the facts callers ask about in one place: hours, services, prices you want quoted, directions.
  • Remove outdated pages, and sync after changes.
  • The agent repeats what it finds, so only add what you want said to callers.