VibeRevert
Beta
AI broke your project? Undo the session, not your week.
VibeRevert records an AI coding session, flags the risky changes with deterministic rules, and restores your project files to exactly how they were before it started, including the work you hadn't committed.
The problem
Your agent changed 84 files. Which ones should worry you?
Somewhere in that diff is a payment route, a migration, or an auth check you never asked it to touch. Your own half-finished work is sitting in the same directory. Reading all of it is not a plan.
Whether you are vibe coding a weekend build or running agents against production code, the failure mode is identical: the agent changed something you never reviewed.
$ viberevert run claude # saves your files' starting state $ viberevert check risk: CRITICAL · payments app/api/checkout/route.ts payments (critical) app/api/webhooks/stripe/route.ts payments (critical) $ viberevert rollback <session> # preview, nothing changed yet package.json tracked_restored app/page.tsx tracked_restored app/api/checkout/route.ts untracked_deleted app/api/webhooks/stripe/route.ts untracked_deleted lib/stripe.ts untracked_deleted .gitignore README.md (your uncommitted work) skipped_unchanged
The last line is the one that matters. Your own uncommitted work was in the same directory the whole time, and rollback leaves it exactly where it was. Add --apply to restore.
Record · Check · Fix · Restore
Four steps around every session.
Capture the starting state
Tracked, staged, and untracked files, including everything you hadn't committed yet.
Let your agent work
No babysitting. VibeRevert records which project files changed while it ran.
Flag risky changes
Payments, auth, database, secrets, infrastructure, dependencies. Each one with the reason it was flagged.
Fix them or restore
Hand your agent a generated fix prompt, or put the whole session back. Rollback previews before it writes.
Determinism
No AI judges the AI.
VibeRevert does not ask another language model whether your agent's changes look risky. Its checks are rules. The same inputs and configuration produce the same findings, every time, with an inspectable reason for every flag.
That is what makes it dependable in a git hook or in CI: the verdict does not drift because a model sampled differently today.
Independence
One recovery layer across your coding agents.
Built-in checkpoints live inside the agent's workflow. VibeRevert stays with the project. Use one agent today and another tomorrow. The safety layer doesn't move out with the tool that created it.
Use VibeRevert with
Proof
Stripe payments in Next.js, a database migration in Laravel, deployment files in FastAPI. Each with our own unfinished work deliberately left uncommitted beforehand, and returned untouched. We also found a real bug in our own preview while doing it, and wrote that down too.
No LLM classifier
Risk is decided by rules, not by a second model. The same inputs and configuration produce the same findings, with an inspectable reason for every flag.
Your pre-session work
Files the agent changed go back to their pre-session state, files it created are removed, and the work you had not committed is left exactly as it was.
8 local tools
A stdio server your agent calls directly, bound to the repository it is working in. No account, no hosted service, no telemetry.
Claude Code, Cursor, Gemini CLI
Tested end to end against real repositories.
Get started
Start with your agent.
First, once per project:
Then choose how you work:
Claude CodePrefer the CLI on its own? npm install -g viberevert@beta
Requires Node.js 22+ and a Git repository. Runs entirely on your machine: no account, no hosted service, no telemetry. Rollback restores your project files; it does not reverse deployments, database writes, or API calls.