Runner Requests

For installing, initializing, status, and lifecycle (agentrunner status/stop/uninstall, supported engines), see Runner Installation. This page covers how to send and track requests.

Send a request

Send a runner request from the web. Set the project, agent, runner, engine/model, plan mode, and RunnerBox option, then enter a prompt. The runner picks up the pending request, executes it via your configured AI agent, and reports the result.

In the prompt, type / to link core workflow items into the request — plans, conventions, co-actions, completion reports, post-mortems, and code reviews (documents and external issues too). The agent picks up that context directly.

Request status

Inspect request details

Open a request from the web runner list to inspect its prompt, status, logs, conversation history, last heartbeat, and RunnerBox state. Running requests can be cancelled, and completed or failed requests can be retried with the same settings.

Click Continue on a completed request to send follow-up instructions in the same conversation.

The follow-up inherits the previous request’s agent, runner, engine, and model, and runs in the same context.

When to use RunnerBox

RunnerBox is the isolated workspace a runner works in (see the Glossary). You can toggle it per request.

When on, the runner creates an isolated workspace from the base branch of the connected repository, so it does not touch the runner machine’s existing checkout. Setting a base branch requires a connected repository. You can inspect RunnerBox state and request its removal from the request detail.

Multi-runner setup

Run multiple runners on different machines, each with its own runner token. Each runner independently polls for and executes requests assigned to its agent configuration.