openronin
A self-hosted AI teammate that works entirely through GitHub: it picks up labeled issues, asks clarifying questions, opens pull requests, replies to review comments, resolves merge conflicts and can auto-merge when checks pass.
openronin treats a labeled GitHub issue like a brief to a remote developer: the agent reads the whole thread, asks clarifying questions when requirements are ambiguous and waits for a reply, then writes code in an isolated working copy of the repository and opens a draft pull request. It replies to review comments in-thread, pushes fix-up commits, marks conversations resolved, and can resolve merge conflicts itself by rebasing and editing conflict markers.
Auto-merge and deploy are opt-in once all comments are addressed and CI is green. The system is split into three layers by role and cost: an optional director service that reads a per-repo charter and proposes what to work on next without editing code, a supervisor that classifies issues, asks questions and plans, and a coder agent that is the only layer allowed to mutate code, confined to an isolated worktree.
The supervisor works with any OpenAI-compatible API, tested against a low-cost model and the Anthropic API, and also supports OpenAI, xAI, DeepSeek, Together, Groq or a local Ollama model. The coder currently uses Claude Code. It is written in TypeScript and MIT licensed.