localhosting:3000

Undo button for localhost. Automatic snapshots when your local app is actually serving a good state. One-click restore of the last good local tree. Nothing leaves this machine.

Not git. Not Vercel rollback. Not a port killer.

You do not remember to snapshot. The watcher does.

MIT, npx, local folder .localhosting/.

Install

Requires Node.js 20.10 or later.

npx localhosting watch

Or from this package:

npm install
npm run build
node dist/cli.js --help

Leave localhosting watch running while your app is on localhost (usually :3000). The dashboard is on :3001 because :3000 is already taken by your app.

npx localhosting list
npx localhosting restore <id>

Restore writes files; restart your dev server. We do not kill processes.

Seatbelt

Stops common destructive commands that pass through localhosting seatbelt run. Defaults on. Fail-closed for beginners. Nothing leaves this machine. A seatbelt, not a vault. Use Rewind to recover.

eval "$(localhosting seatbelt wrap)"
localhosting seatbelt run -- rm -rf dist
localhosting seatbelt status

Seatbelt is a seatbelt, not a vault. Best-effort guard against common mistakes on localhost. It cannot stop every destructive command, kernel or hardware failure, or malicious code you chose to run. Commands that do not pass through seatbelt run are unchecked. Keep real backups. Use Rewind snapshots for recovery.

How to verify no network

v1 is loopback only. There is no telemetry, no accounts, and no postinstall script.

1. Confirm there is no postinstall: npm pkg get scripts.postinstall prints {}.
2. The process may listen on 127.0.0.1:3001 through 3010 and may GET the configured loopback probe URL (default http://127.0.0.1:3000/). That is the entire network surface.
3. Automated tests spy on fetch / http.request during snapshot and expect zero outbound connections.

Commands

init, watch, serve, snapshot, list, restore, pin, unpin, prune, status, seatbelt

Limitations

localhosting:3000 does not promise:

1. Durability beyond this disk. The store is files in .localhosting/. If the disk dies, the snapshots die.
2. A snapshot of files it ignored (by default: node_modules, .git, build artifacts, most gitignored paths).
3. A snapshot during a stretch where the detector never saw a good state (overlay stayed up, probe URL down, settle never happened).
4. Preservation of in-memory dev-server state, HMR runtime, cookies, or browser tabs.
5. Cross-machine restore, encryption-at-rest beyond OS permissions, or multi-user locking beyond a pid lockfile.
6. That the latest snapshot is the user's favorite state. It is the latest detector-good state. Pin exists for favorites.
7. Database undo (Postgres, SQLite files that were gitignored, Redis).
8. That a restore will succeed on every open file while another process holds a Windows lock. Partial restore is a specified outcome (exit 4).

git is not involved. We do not touch the index. Restored files may appear as git diffs. That is correct.

Do not give an agent localhosting restore --yes in a skill file. Dashboard restore always confirms.

Never pass --yes to an agent. --yes skips the uninstall confirm and is for Diet's own test suite.

Not a port dashboard

We do not scan your ports. We do not kill processes. Discovering a free bind port for our own dashboard (3001 then 3002-3010) is bind-and-retry, not a scanner UI.

FAQ

Why not just use git? You would. You did not. The agent did not either. This watches the running app, not your conscience.

Is this a production rollback tool? No. Production already has Netlify and Vercel. This is for the three hours before you deploy, when :3000 is on fire and there is no commit.

Will this replace Time Machine / uhoh / local history? No. Those snapshot writes. We snapshot good running states. Different trigger, smaller history, on purpose.

Do you upload my code? No. v1 has no accounts and no telemetry. Snapshots live in .localhosting/ on this disk. If the disk dies, they die.

Can I restore without restarting the dev server? We write files. We do not kill processes. Restart the server. If a file was locked, restore reports partial and you retry.

Where is the dashboard? :3001 by default, never :3000. :3000 is the patient, not the doctor.

License

MIT