supervisor-api-background-mode

To run long-running agent tasks that exceed HTTP timeout limits, use Supervisor API background mode to asynchronously submit requests, poll for completion, and stream results back to the frontend. This provides resilience against connection drops during execution.

Background mode handles asynchronous requests by submitting them to the Supervisor API, polling for completion at configured intervals, and streaming results back to the frontend. This mechanism ensures complex multi-tool workflows complete successfully beyond HTTP timeout limits.

Yes, you need the AsyncDatabricksOpenAI client to use Supervisor API background mode. It is required to manage asynchronous requests, handle multi-turn approval flow for MCP tools, and maintain resilience to connection drops during long-running agent tasks.

The best way to set up the Supervisor API with hosted tools and permissions is by running `uv run quickstart`. You can then configure the poll interval in `agent_server/utils.py` to manage how frequently the system checks for task completion.

Agent tasks fail during connection drops because standard HTTP requests are limited by timeout constraints and cannot maintain state. Supervisor API background mode solves this by asynchronously submitting requests and polling for completion, ensuring resilience to connection drops.