Knowledge Vault (RAG)

Knowledge Vault (RAG) retrieves knowledge passages for the next step in the graph.

Purpose and use: Exposes configured knowledge operations to an AI Agent; use when the model should choose a permitted knowledge query or document operation. This provider is distinct from the direct Knowledge (RAG) action node.

Configuration: Optional selectedSources (string array); allowedOperations (default ["kb_search"], supported values include kb_search, rag_search, rag_document_upsert, rag_document_update, rag_document_delete); and knowledgeScope object. Other operations require explicit grants.

Input/output and routing: Input input; outputs output and tool_out. Connect tool_out to AI Agent tools_in. It provides capabilities rather than executing a normal control-flow step.

Expressions: Source and operation grants are static. Agent prompt expressions can refer to results after a provider tool is invoked; those results are specific to the invoked operation.

Errors, permissions, and effects: Empty/unknown grants expose no tool. Knowledge scope is validated and intersected with workflow/binding grants. Search is read-oriented; document upsert/update/delete are writes and may require approval. Confirm sources, scope, audience ceilings, operation grants, and approval before enabling.

Example configuration — fields validated against the 1.1.1 node schema; replace the identifiers with your own: {"selectedSources":[],"allowedOperations":["kb_search"]}. Expected result: only the search capability you configured, exposed to the connected agent.