Skip to main content
This page uses the original routes, which keep working. For new integrations, see Search. Start the search here, then poll, download, and page through it with the Search Results endpoints. For the end-to-end flow, see the Deep Search Overview.

API Reference

See the full request/response schema and parameters in the API Reference.
Submit a natural language query. Returns a task_id to poll with GET /search/status/{task_id}.
Response:
The task ID is agentic_search_ followed by the search’s search_id.

Concurrency and run time

Each organization runs up to 5 deep searches at a time; see Concurrency limits. A deep search runs for at most 15 minutes, then ends timed_out.

Retry a start request

A start request that sends its own search_id (a UUID) can be checked after a timeout at GET /search/status/agentic_search_{search_id}:
  • A status response means the search started.
  • A 404 two minutes after the start means it did not start. A new start needs a new search_id.
Each search_id must be a new UUID; see Search errors.

Get the results


Example usage

Rate limits

For per-endpoint limits, see Rate Limits.

Pricing

See Credit Costs for more information.