grove

If you run a lot of jobs against the same repo, you know the drill. Agents, CI, parallel tasks, whatever. Clone again. Fetch again. Wait. Grove is what I built to get around that.

It's a TypeScript SDK and CLI that keeps a pool of Git worktrees ready to go. You acquire a lease, get an isolated checkout tied to a stable leaseId, do your work, and release when you're done. No full re-clone every time. No sitting there watching git fetch while three other jobs pile up behind it.

Check the code on GitHub

What problem it solves

Most tools treat every job like a fresh start. Grove treats checkouts more like something you borrow and bring back.

You reserve a slot with a lease ID, check out a branch or ref, run your thing (even if the process restarts), and then release, repair, or destroy when you're finished. Your commits and dirty state stick around until you pick a cleanup policy: keep it (preserve), reset it (reset), or quarantine it if something went sideways.

What I like about it

A few things I'm especially proud of:

• Durable leases. State survives restarts. Re-acquire the same leaseId and pick up where you left off.
• Fast reuse. Reset uses git clean -fd, not -xfd, so ignored stuff like node_modules stays put between runs.
• Crash recovery. Write-ahead state and an explicit repair() when something gets stuck mid-operation.

CLI and SDK

Grove ships two ways to use it.

There's a CLI (@ferueda/grove-cli) when you want to run commands from a terminal or shell scripts. Every command supports --json with stable envelopes, which is great for glue code and quick automation.

I also built a TypeScript SDK (@ferueda/grove) because my agents and automation needed to spin up worktrees directly in code. No shelling out to a CLI, no parsing stdout. Your orchestrator calls acquire, gets a path back, runs its work, and calls release. That's the interface I actually wanted for agent workflows.