Create a long-running document agent job. Returns immediately with a job_id — use GET /document/agent/jobs/ to poll for the result or GET /document/agent/jobs//stream for live SSE updates.
Recommended for production workloads, batch evaluation pipelines, or anything that may run longer than ~10 minutes. The job runs on a background worker pool with 30-minute hard timeout, per-user concurrency caps, and automatic refunds on failure.
API key must be provided in the Authorization header
Request for the v2 document/agent endpoint.
Question to ask about the document(s)
1 - 10000Data source ID of a single indexed document
List of data source IDs for multi-document queries (max 10)
10JSON Schema for structured output
Model to use (claude-opus-4-6-1m, claude-opus-4-6, claude-sonnet-4-5-20250929, etc.)
Enable extended thinking
Token budget for thinking (ignored for adaptive models like Opus 4.6)
1000 <= x <= 50000Stream response as SSE events