What is forge?

Forge is a local-first tool for people who run AI coding agents like Claude Code, Codex, or Gemini and want a proper way to manage the work those agents produce. Instead of letting an agent edit files directly in your main project folder, Forge gives each task its own isolated git worktree, so agents cannot interfere with each other and you can discard a task's changes without touching the rest of your repository. Every task moves through a clear lifecycle: from to-do, to in progress, to review, to merging, and finally done, with an audit log tracking what happened along the way. You can define CI steps for each task, and Forge will block the work from merging until those checks pass. This gives teams a review gate similar to what they would expect from a human pull request process, but applied to agent-generated code. The project runs entirely on your own machine. It uses a single binary and SQLite for storage, with a local server that only listens on your own computer, so there are no cloud accounts, no telemetry, and no outside service required. You can install it through npm, Homebrew, or a shell script, and there is a demo mode that seeds sample data so you can try the interface before connecting a real agent. Forge supports multiple ways of interacting with it: a REST API, an MCP endpoint, a command line tool called forge-ctl, and a built-in web interface built with React, all shipped in the same binary. It is built specifically for people who already use coding agents and want a control layer around them. It is not meant to be a chat interface bolted onto an editor, if that is what you want, the project itself points to alternatives like Cursor or Cline instead. The project is currently in public beta, meaning parts of its API and command line flags may still change before a stable release. It is released under the MIT license.