Skip to documentation

Features

Code Agent

Implement eligible GitHub Project work and return a verified pull request.

From Ready issue to pull request

Code Agent runs for eligible GitHub Project issues that move to Ready. It claims the work, moves project state forward, creates or reuses a preserved workspace, implements the change, validates it, and opens a pull request.

A successful run requires one verified pull request with the expected branch, commit, and evidence markers.

Configure the implementation role

The configured image is prepared by the connected runner and applies to later jobs. A running pod keeps the image snapshot it started with.

  • Pause or enable the agent independently.
  • Choose a provider, live catalog model, and supported thinking level.
  • Add repository-specific instructions.
  • Use the Orchestrator default image, a trusted local image, or a registry image.
  • Control shared repository capacity from repository settings.

Preserved implementation workspace

Pull-request artifact work uses a preserved repository workspace so retries and later attempts can continue from durable local state. Events and GitHub evidence remain attached to the run.

Frequently asked questions

What triggers Code Agent?

An eligible issue in the configured GitHub Project becomes Ready and satisfies repository capacity and runner readiness.

What must a successful run create?

One verified open pull request with the expected branch, commit, and run evidence.

Can Code Agent use a custom image?

Yes. Select a tagged or digest-pinned local or registry image that satisfies the agent contract.

What happens when implementation is blocked?

The run records a blocker or failure with evidence instead of reporting a successful pull request.

Do provider credentials reach the hosted website?

No. Provider API keys and OAuth roots remain in the local runner environment. The hosted control plane receives sanitized availability, model catalog metadata, job events, and outcome evidence.

Can agents in one repository use different models?

Yes. Each built-in or custom agent selects its own provider, catalog model, supported thinking level, and repository tool image.