long-running-server

You prevent long-running AI agent requests from timing out by submitting work asynchronously with background=true, then retrieving results later via GET /responses/{id}. This allows background tasks to complete reliably beyond standard HTTP timeout limits.

Streaming with resumption works by streaming incremental events for long tasks and supporting cursor-based continuation via starting_after. This allows clients to resume streaming background agent workloads after network interruptions or timeouts.

Yes, you need Lakebase PostgreSQL to persist long-running task state. Lakebase persistence ensures clients can poll or resume background agent execution after timeouts by storing task state reliably in the PostgreSQL database.

Yes, you can use background polling for multi-step reasoning and tool calls. Configure LongRunningAgentServer with task_timeout_seconds and poll_interval_seconds to ensure reliable background processing for complex agent workloads over lakehouse data.

Polling intervals for asynchronous agent execution are limited by configurable poll_interval_seconds and task_timeout_seconds settings. If background work exceeds task_timeout_seconds, requests may fail, so configure these parameters based on your workload's complexity.