Create Agent Run
Start an agent run. The run executes asynchronously: the response returns immediately with status queued, then poll GET .../runs/{run_id} until completed and fetch the output from GET .../runs/{run_id}/result — or set enable_events: true and follow GET .../runs/{run_id}/events for live progress.
To enrich existing records instead of researching from scratch, pass them in input_data; this requires an output_schema (on the request or the agent).
Authorizations
Body
JSON Schema overriding the agent's default structured output for this run. Root must be an object with non-empty properties or an array of objects. Hard limits: nesting depth ≤ 5, ≤ 100 properties, ≤ 500 enum values, no unsupported keywords (format, pattern, min/max*, root anyOf, standalone null). Invalid schemas return 422.
Only settable when this run creates a new agent (via agent_name, or when no agent is resolved), in which case it becomes the new agent's stored use_case. For a run against an existing agent, this must match the agent's own use_case - passing the same value is accepted as a no-op, a different value is rejected.
research, enrichment, dataset_building Response
Successful Response