You built it.Now put it liveby sending a message.
You made something — probably with an AI that wrote most of the code — and it works on your laptop. It is on GitHub. You have bought a server. And now the instructions say things like “configure a reverse proxy” and “set up a systemd unit”, and that is where it has sat for three weeks.
Connect the server to WaveSenda and you can skip that part. Ask it questions in WhatsApp. Tell it to deploy. If the deploy breaks the site, it puts the old one back before it tells you.
What it looks like
A conversation on the number you already use, with an agent that is only allowed to do what you ticked.
What it is allowed to do
This is the question anybody sensible asks within half a minute of understanding the idea, so it is answered here rather than buried in the documentation. An agent with a shell on your production server is a bad idea. This is not that.
Looking
Runs straight away
Service status, log tails, disk and memory, open ports, certificate expiry, what was deployed and when. These change nothing, so they need no confirmation — a chat that asks permission to check whether your site is up is a chat nobody opens twice.
Changing
Asks first, and can be undone
Restarting a service, deploying, renewing a certificate, clearing space. Each one asks you to confirm in the thread and then asks for a code sent to your email — a second factor on a different device, because somebody who has taken over your phone has your WhatsApp. Every one records how to reverse itself before it runs.
Cannot be undone
Never offered in a chat
Deleting a database, removing a volume, changing the firewall, editing who can log in. These are not “ask twice” actions. They are not in the list the agent is given, at any level of insistence, and they are done from the console where you can read what you are about to do.
That last rule exists because of who this is for. The usual safeguard for a risky action is to ask a human first — but somebody who knew that dropping a table was the wrong move would not need this product. Asking them to approve it only moves the blame. So the category is removed instead.
Deploys that put themselves back
Deployment is the reason most people arrive here, and it is also the action most likely to take a working site down. So it is built the careful way round.
You never learn what a symlink is. You type deploy, and either it worked or your site is already back on the version that did.
- 01Each deploy builds into its own directory alongside the version that is running.
- 02Switching over is a single atomic step, so there is no moment where the site is half-deployed.
- 03A health check runs immediately afterwards: your site has to answer before the deploy counts as finished.
- 04If it does not, the previous version is put back automatically and you are told that is what happened.
- 05The old release stays on disk, so “undo” is instant for the next hour.
How it reaches your code — and your server
Your code
Through a GitHub App you install on the repositories you choose. It asks for read access to your code and nothing more — an agent that deploys does not need to be able to change what it deploys.
When something lands on your main branch, the agent tells you what arrived and offers to put it live. It does not deploy on its own: automatic deployment on push is a feature for people who have tests.
Your server
The obvious way to build this is to store an SSH key for every customer and connect in. It is also the way that makes a single break-in catastrophic for everybody at once, so we did not.
Instead a small agent runs on your machine and dials out to us. Nothing connects in. No port is opened, no firewall rule changes, and it works behind NAT. The list of things that can be done to the server is compiled into that agent, on your machine — so even if our own systems were compromised, what could be asked for would still be limited to what you installed and ticked.
The agent is open source and published with a checksum, so you can read every line of it before you run it. It is here.
Agents for the different jobs
Deployer
Puts your code live, checks it worked, undoes it if it did not
Watchman
Tells you the disk is filling or a certificate expires in six days — before you notice
Medic
“The site is down.” Finds out why, fixes the smallest thing that fixes it
Warden
What is exposed to the internet, what needs patching. Mostly advice
Caretaker
Backups, log rotation, certificate renewal, tidying up
Concierge
For somebody who has never run a server. Explains everything as it goes
Common questions
Do I need to know Linux?
No. That is the point of it. You connect the server once by pasting a line your provider’s web console will accept, and after that you ask in plain words — “is my site up”, “deploy the latest”, “why is it slow”. The agent explains what it found and what it is about to do before it does anything.
What can it actually do to my server?
Exactly what you tick, and nothing else. A newly connected server can only look at things: service status, logs, disk space, certificates. Restarting, deploying and renewing certificates are switched on one at a time, and each asks you to confirm. Deleting data, changing the firewall and touching SSH keys are never offered over WhatsApp at all.
What happens if a deploy breaks my site?
It puts the old version back on its own. Every deploy builds alongside the running one, switches over, and then checks that your site still answers. If the check fails, the previous version is live again before you are told anything went wrong. You can also reply “undo” within the hour after any change.
Do you have a key to my server?
No. A small agent runs on your machine and dials out to us — nothing connects in, no port is opened, and we never hold a credential that opens a shell. The list of things that can be done is compiled into the agent on your machine, so it is not something we can widen from our side.
Which servers does it work with?
Any Linux server with systemd that you can run one command on: Hetzner, DigitalOcean, Contabo, Vultr, Linode, OVH, AWS Lightsail, a machine under your desk. It is not for managed hosting where you cannot install anything.
What does it cost?
It is part of your plan rather than an extra subscription. Every plan can connect a server and ask it questions; making changes is included from Pro upward. Asking questions uses AI credits like any other agent reply — the actions themselves cost nothing, because the work happens on your own machine.
One line on the server, and it is connected.
Paste it into the web console your provider already gave you. Nothing else to install, nothing to open.