Skip to main content
A knowledge base lets an agent answer from your content instead of guessing. Documents are embedded and stored for vector search. During a call the agent retrieves the most relevant entries and uses them to ground its reply.

​
The model

A knowledge base is a named entity in your workspace, created and filled independently of any agent. An agent points at one through knowledge_base_id:
  • One knowledge base per agent. An agent retrieves from exactly the KB it is bound to — nothing is shared implicitly across KBs.
  • Many agents per knowledge base. Several agents can share one KB, so the same content doesn’t need re-ingesting per agent.
  • The agent applies a confidence threshold (default 80%) and ignores low-scoring matches.
  • Names are unique within a workspace; creating a duplicate returns 409.

​
Manage knowledge bases

​
Manage documents

Ingest and search are scoped by knowledge_base_id, not agent_id. Retrieval is exact-per-KB: searching one knowledge base never returns another’s entries.

​
Search shape

POST /knowledge-base/search returns ranked Q&A matches:
Use match_percentage to decide how much to trust a result; the agent already filters below its threshold during live calls.
The knowledge base requires a Vector Store to be configured on the deployment. When it isn’t, KB features are disabled and search returns a service-unavailable response — calls still work, just without retrieval.

Agents

Attach a KB to an agent.

Sessions

Retrieval happens mid-call.