Hi everyone, I'm Hung. I built Coppice because running several coding agents in different terminals made me a worse reviewer, not faster. Five tabs, each one half-done, and I kept losing track of which agent was doing what and what I had actually read.
Coppice is a desktop app with a board, and it drives the CLI agents you already pay for: Claude Code, Codex, Cursor, OpenCode, Kilo Code and a few more. No new subscription, no account, no Docker. It uses your CLIs signed in the way they already are.
Every agent gets a role: a PM agent plans, others build, review and test. You're the lead, and there are two gates where you have to say yes:
You write a ticket and a PM agent writes the plan. Nothing gets built until you approve it. It can ask you a couple of questions first if the ticket is vague. (For small stuff you can skip planning.)
Agents build, each one in its own git worktree, so your checkout isn't touched. Other agents review and test it. Then the diff comes back to you in the app. You comment on a line, the same agent picks it up and fix it on the same branch. Accept is not merge. Merging is its own step, and what you accepted is exactly what gets merged.
Each role can use a different CLI and model, like a stronger model for planning and review and a faster one for coding. Agents only talk to each other through ticket comments, so if something happened, it's on the card.
It's open source (Apache 2.0), for macOS and Linux (Windows via WSL).
Heads up, this is a beta. It works for my daily work, but you will probably hit rough edges, and some CLIs are tested less than others. So thank you in advance if you try it. Any feedback helps a lot, even "it didn't install" or "I didn't get it". Issues are here: https://github.com/fertile-org/coppice/issues
What I'd love to hear most: does the plan gate feel useful or just in the way, and what would stop you from trying it?
Docs: https://getcoppice.app
Learn more:
https://github.com/fertile-org/coppice
Top comments (0)