berth

git worktree copies your code. Berth copies everything else.

Each worktree gets its own ports, hostname, .env, dependencies and database. Run npm run dev, rails server or docker compose up in all of them at once.

Any language · plain dev servers · local MySQL or Postgres · Docker Compose

Two worktrees fight over five things

A worktree copies tracked files and nothing else. Run two at once and they collide, so you end up testing one branch at a time.

  • PortsBoth dev servers want port 3000.
  • Names and cookiesLogging in on one worktree logs you out of the other.
  • DatabasesA migration on one branch breaks another branch's tests.
  • Untracked filesA new worktree has no .env and no node_modules/.
  • The one local databaseEvery worktree points at the same Postgres or MySQL.

Berth gives each worktree its own. You start things exactly as you do in main.

Four agents, four running apps

Agents already work in worktrees. With berth, each one can run, migrate and test its branch for real, without waiting its turn or breaking a neighbour's data.

feat-couponsRails · local Postgres
$ bin/rails db:migrate
== AddCoupons: migrated
$ bin/rails test
214 runs, 0 failures
new-billingNode · dev server
$ npm run dev
VITE ready on :20003
$ berth status --json | jq .url
"http://new-billing.localhost:20002"
fix-checkoutDocker Compose
$ docker compose up -d
[+] Running 4/4 shop-fix-checkout
$ curl -s localhost:20000/health
ok
drop-legacyDjango · local MySQL
$ python manage.py flush --no-input
Flushed shop__drop_legacy
main's database is untouched

Agents run berth status --json to find their own URL and ports.

Works wherever you make worktrees

Berth works on git worktrees themselves, so it doesn't matter which tool made one. Install berth's git hook once and anything that runs git worktree add gets a full environment, with no plugin, wrapper or per-tool setup.

  • Your terminal
  • Claude Code
  • Cursor
  • Codex
  • Conductor
  • Any script
git worktree addthen berth add, or berth's git hook does it for you
berthports, names, files, data

A new worktree is ready in seconds

No reinstall. No dump and restore. berth add clones what your branch needs and hands the terminal back while the big copies finish in the background.

InstantPorts, names and .env

Assigned and written the moment you run berth add, however large the project.

0.5 snode_modules/ and vendor/

30,000 files cloned copy-on-write, so there is nothing to reinstall and almost no extra disk used.

1.6 sA 1 GB Postgres database

Cloned inside the server on Postgres 18. Your branch gets real data to migrate, and main's stays as it was.

See how berth copies each kind of data

And safe with your data. Berth never starts your services, never changes main's data, and deletes only what it created. Read the guarantees

Set it up once, then forget it's there

One binary, one command per repo. Your app, your scripts and your compose file stay exactly as they are.

$ curl -fsSL https://useberth.dev/install | sh
Once per repo
$ berth init
When a worktree needs to run
$ git worktree add .worktrees/x
$ cd .worktrees/x && berth add
$ npm run dev
Or make it automatic for every worktree
$ berth hook install