Wit is an agent coordination protocol that prevents merge conflicts when multiple AI coding agents work on the same repository simultaneously.
The problem Wit solves is that multiple AI agents working independently on shared code produce merge conflicts that waste time and break work. Git detects these conflicts after they happen. Wit prevents them before code is written by running a lightweight daemon that tracks four primitives: intents (announced planned work scope), locks (reserved functions, types, or classes), conflicts (warnings when intents or locks overlap), and contracts (agreed function signatures). Agents communicate with the daemon over a Unix socket using JSON-RPC. Intents and locks serve as warnings rather than hard blocks, allowing agents to make their own decisions about how to proceed. Contracts are the only hard enforcement mechanism, applied through a git pre-commit hook that blocks commits violating accepted signatures.
Wit suits teams running multiple Claude Code instances, Cursor, Copilot, Devin, or other AI coding agents on the same repository. It works by having agents declare their work scope upfront and lock specific symbols before writing code, enabling the daemon to warn about overlapping work. The tool requires Bun for its daemon, SQLite, and process management. After initialization with the `wit` command, a `.wit/` directory and `CLAUDE.md` file are created; subsequent Claude Code sessions automatically read these coordination instructions without manual setup.
The project maintains active continuous integration. Development activity shows consistent engagement with the codebase through regular updates and testing infrastructure.
