Ringer
A local coding agent orchestrator that runs parallel CLI workers, checks their output and tracks progress in a browser dashboard.
Ringer coordinates parallel AI coding agents on your machine. It's for developers who want to divide implementation work among workers and judge the results with executable checks.
- Executed checks: Ringer runs a supplied check command against each task's output and retries failed tasks once with the failure details.
- Worker selection: Tasks can use different engines and models, including Codex CLI, Grok Build CLI and OpenCode with OpenRouter, so planning and implementation can use separate models.
- Live dashboard: Ringside is a local browser page that shows active runs, worker activity, elapsed time and token use, plus a versioned library of earlier outputs.
- Performance records: Local attempt logs track check results, duration, tokens and model identity, letting developers compare first-attempt success rates by model and task type.
Codex CLI is the default worker. Other headless command-line workers can connect, but model access depends on the chosen engine. The OpenCode integration uses models served through OpenRouter, so that route sends model requests to an external service. Ringer stores its attempt logs locally by default.
The supplied OpenCode sandbox confines writes on macOS while allowing network access and reads. Other platforms need a separate sandbox or full-access mode. A passing result means the supplied check passed; the strength of that check determines what it establishes.
