Let your agent ship.
Replay proves it.
Rewind undoes it.
Agents write code faster than anyone can review it. Instead of slowing them down, velixir puts the guardrails in the platform: every release is tested against your real traffic before it goes live, and your whole app can go back to any moment if something still gets through.
What comes back
Writing code got fast. Checking it didn't.
An agent can open more pull requests in an afternoon than a team can read in a week. The usual answers are to read every line, which brings you back to human speed, or to merge and hope, which is how production databases get deleted. Neither is good enough.
So we moved the safety net to where it can do the most: the platform your app runs on. It already sees every request, every release and every backup. Rewind and Replay put that to work.
Proven against real traffic before it's live.
Tests check what someone thought to write down. Replay checks what your users actually do: it sends your recent real requests to the new release and to the one that's live, and shows you every difference.
Real requests. It picks your most visited pages and API calls from your access logs, plus a sample of the rarer ones. No traffic yet? It follows the links from your home page instead.
Two sealed copies. It makes two copies of your app from the same moment, each with its own copy of your data: one on the live release, one on the new one. Neither can reach the internet, so nothing replayed touches production or the outside world.
Only the real differences. The live release answers every request twice, so anything that changes on its own (timestamps, generated ids) is ignored. What's left is new errors, changed status codes, changed responses and slower routes.
Your agent can read the report over MCP and fix what it finds before a person ever looks.
Your whole app, back to any moment.
Most rollbacks bring back the code and leave you to sort out the rest. Rewind brings back the code, the env vars it ran with and the data it used, together, to the same moment. Or only the parts you choose: just the code after a bad deploy, code and files after a broken upload, everything after an agent dropped a table.
Pick a moment and the parts. Drag the timeline. Each part tells you exactly what it can come back as before you commit to anything.
Look at it first. You get a preview of the app as it was, on its own URL, with its own copies of the data. Nothing changes on the live app yet.
Apply, and undo if you change your mind. The live app moves onto the restored copies. The original databases and caches keep running untouched, so going back is one more click.
How exact each part is
Shown on the timeline before you rewind, so there are no surprises.
Rewind restores your app, not the outside world: emails already sent and payments already taken stay sent and taken.
Your agent can use the safety net itself.
Rewind, Replay and forks are tools in our MCP server and commands in the CLI, so an agent can test its own release, give itself a copy of production to work on, and hand you a rewind to approve.
Originals never overwritten
A rewind restores into new copies and points your app at them. The databases and caches you had keep running, untouched, which is what makes undo possible.
Copies can't reach the internet
A copy has your env vars, live API keys included. Previews and Replay copies are sealed: they reach their own data and nothing outside, so a background job in one can't charge a card or email a customer.
You approve what goes live
An agent can build a preview and test a release on its own. Making a rewind live is a separate step, and our MCP tools tell agents to ask you first.
Rewind + Replay questions
What does it cost?
Rewind and Replay themselves are included. The copies they make are normal apps and databases, billed by the hour like any other while they run. Replay discards its copies when it finishes; you discard a preview or a fork when you're done with it.
How far back can I go?
Code and config for 28 days. Databases as far as your database plan keeps backups: 7, 14 or 28 days. Caches and files as far back as their snapshots go.
Can I rewind only part of my app?
Yes. Leave out the code, the config, the databases, the caches or the files, or a single database or cache. Anything you leave out stays exactly as it is now.
What does Replay send?
Recent GET and HEAD requests from your access logs, the ones that shouldn't change data. It reads up to the last 7 days on paid plans and the last day on the Free plan.
Will a copy send emails or charge cards?
No. Previews and Replay copies can't reach the internet at all. A fork you make yourself is sealed too, unless you choose otherwise.
Does it work on the Free plan?
A copy is an app on the same plan as yours, and the Free plan includes one app. To rewind, fork or replay a Free app, move it to a paid plan first.
What can't it undo?
Anything outside your app: an email that was sent, a payment that was taken, a webhook another service received. Rewind restores your app, not the world around it.