Ringer
Local AI agent orchestrator for macOS, Linux and Windows via WSL. Runs parallel workers through Codex, Grok Build or OpenCode with OpenRouter.
Ringer runs parallel AI coding agents on your machine and checks their output by executing tests or other validation commands. It's for developers who want to delegate implementation while keeping planning and review with a stronger model. The aim is to reduce model spending on routine coding work without relying on workers' claims that they're finished.
Each worker gets a separate task directory, with optional Git worktrees to keep concurrent edits apart. Failed tasks can retry with the check's failure output. Ringer also examines task definitions for weak checks and conflicting writes, and can test checks against unchanged code before any workers run. Passing a check doesn't establish that a change is logically correct, so code review still matters.
Ringside, its local browser dashboard, shows active batches, worker progress, deliverables and check results. It keeps a versioned library of past outputs. Local attempt logs support comparisons of model pass rates, duration and token use by task type, including which tasks needed retries.
Ringer runs on macOS and Linux, plus Windows through WSL. It supports Codex CLI, Grok Build CLI and OpenCode with OpenRouter, and can connect to other headless command-line workers. The orchestrator, dashboard and run records stay local; the named cloud worker services require authentication and send model requests off the machine. The repository uses the PolyForm Shield license.
For questions about supplied files, a separate read-only mode selects relevant passages for a single worker. That mode checks that an answer exists, not that its claims are correct.
