Put an app online

From "I have an idea" to "here is a link you can open" — the whole path, in order, with the parts that will stop you named up front.

You do not need to be a developer to use this. You do need about an hour, and you need to know which parts are still genuinely hard — they are named below rather than discovered at 11pm.

If you have used something like Lovable or Bolt, the difference is this: those build and host in one place. Here, Claude builds and OMMAX hosts, which is why there are two connections to make instead of none. In exchange the thing you build sits behind the company login and can use company data.

The short version

1. Register the app — reserves the name, the web address and who may see it.
2. Build it — Claude Code writes the app on your machine.
3. Put the code somewhere the platform can read — a git repository.
4. Connect your deployment tools once — one click on the connect page.
5. Ask Claude to deploy it — the address from step 1 starts serving it.

Steps 1, 4 and 5 take minutes. Step 2 is the fun part. Step 3 is the one that will stop you if you have never used git — read that section before you start.

1 · Register the app

Launcher → Create an app. Give it a name and describe what it does in a sentence or two; that description is not decoration, it is what colleagues see while the app is still being built.

You get back a web address immediately. It works right away and shows a placeholder saying the idea is registered but not built yet — that is expected, not a failure.

Who can see it: if you are a team lead or above, it appears on the launcher straight away. Otherwise it stays private to you and one person you name as approver until they make it visible.

2 · Build it with Claude Code

Describe the app to Claude Code the way you would describe it to a contractor: what it does, who uses it, what it should look like, what data it shows. Ask for something small that works before asking for everything you want. You will go back and forth a few times — that is normal and it is the whole point.

Two things to ask for explicitly, because they matter later and Claude will not guess them:

• "Make it run in a container" — a Dockerfile. Hosting builds from that.
• "Make it listen on a port and tell me which one" — you will need the number.

3 · Put the code where the platform can read it

This is the honest wall. The platform builds from a git repository — never from the files on your laptop — and it cannot create that repository for you. Today it can only read repositories in the company Bitbucket workspace.

So: ask Claude Code to set up git and push, and ask someone in Engineering for a Bitbucket repository if you do not have one. It is one message, and it is the one part of this journey that still needs another human.

Being fixed: per-app repository access so you can bring your own repository with a token instead of needing that workspace. Until then, this step needs a colleague.

4 · Connect your deployment tools

Avatar menu → Connect your tools (MCP) → section Connect your deployment tools → Connect this host. Copy the block it gives you into Claude Desktop's config, restart it, done.

You never see or type a hosting password — the platform issues your own credential behind the scenes and hands your AI host a token. That token is yours and carries your own permissions: you can do exactly what you could do by hand, and nothing more.

Not using an AI host? The same page has "Not using an AI host? Get an API key for scripts and curl" — same access, a plain key.

One token at a time. Connecting again issues a new token and kills the old one, so if you set up two machines the first stops working until you paste the new token there too.

5 · Ask Claude to deploy it

Now just say it: "deploy this app to the address I registered". Claude will create the service, point it at your repository, set the build, attach the address, and release it. It will ask you to confirm before the deploy itself — that is deliberate; deploying is immediate and public.

When it finishes, open the address from step 1. The placeholder is gone and your app is there.

The hosting console opens with your OMMAX login

Your app sits in a hosting project of your own, and you can open the platform's own console to look at it: your launcher's hosting tile, Dokploy → Prod. It signs you in with your normal OMMAX login — the same one that got you into Atrium — and lands you on your projects. There is no hosting password to type, and nobody will send you one.

The credential from step 4 is the machine's door, not yours. It is what lets your AI host act as you over the API; it was never a login you were meant to use, which is why you were never told it.

Your seat has to exist first. If the console says you have no hosting access provisioned, you do not have a project there yet — that is step 4. No password was rejected, because there is no password in this flow at all.

Console or tool call, same projects. Asking is usually faster:

What the tools do not cover — databases, backups, and deleting anything at all — is console work. What you may actually do there is whatever your seat permits; if the console refuses you, that is a real limit and a message in Report a problem, not something to route around.

When something goes wrong

What this costs you to maintain

An app that is online is yours to look after. You can read its logs and redeploy it by asking Claude. Nobody else is watching it. If it stops mattering, say so — leaving it running is not free, and there is currently no automatic clean-up.