FAQ
The questions people actually ask — about what agenthost can host, what happens when things break, who can see what, and where your data lives.
Getting started
No. You need an agent that can write the app and a rough idea of what you want it to do. It helps you nothing to know what a container is — agenthost has no settings for any of that, for anyone.
Anything that speaks MCP: Claude (app or claude.ai), Claude Code, Cursor, VS Code with Copilot agent mode, and many others. See Connect your agent.
No, and that's deliberate. Everything agenthost does is a tool your agent can call, so asking is the interface. If you want to see what you have, ask: "list everything I have on agenthost."
No. Every app gets a free address with HTTPS. Bring your own domain when you want one — see Use your own domain.
The MCP server is the API. Anything that speaks MCP can drive it, including scripts and CI — see Deploy from CI. There is no separate REST API and no CLI.
Building and deploying
Static sites, and apps in Node, Python or PHP. If the runtime conventions don't quite fit, commit an
agenthost.yaml to control the install, build and start commands.
Anything those four runtimes can't cover isn't supported — see Runtimes.
Yes. Ask your agent for a database and it provisions a managed PostgreSQL, MySQL or Redis and wires it into your app — the connection string, password and all, arrives as an environment variable your code reads. You never see or paste a password, and there is no separate database to buy — one is included in what an app costs. See Databases.
Databases are opt-in, never created automatically, so an app that doesn't need one doesn't get one.
An app can equally talk to a database you host elsewhere — put that connection string in as an
environment variable yourself, at /secrets-manager, so the password
never goes through your agent.
Deploy again. Each deploy replaces the last, so shipping a fix is the same request as shipping the change. There is no rollback tool, though — keep your project in git (or just keep the files) and your agent can redeploy an earlier copy. See Update what's live.
Ask your agent. It can read your app's build and runtime logs directly and usually explains what happened in plain language, with the fix. See Fix a broken deploy.
Usually under a minute. Static sites are fastest — there's nothing to build. An app with a lot of dependencies to install takes longer on its first deploy and less afterwards, because dependency layers are cached.
Not directly — an app's hostname is fixed when it's created. Your agent can deploy it again as a new app on the new address and delete the old one. See Use your own domain.
Access and privacy
Yes, two ways, and neither is a login for you to build. Name the email addresses allowed in and everyone else meets a sign-in screen: your visitors get in with an emailed code, with no account to create. Or set one shared password and give it to whoever should see the thing. Invitations are the stronger lock, since they tell you who has opened it and let you remove one person without disturbing anyone else. See Make something private.
No. Environment variable values are write-only: they go in when the app is created and are never returned by any tool. Only their names are readable. Your agent also never sees your agenthost credentials — sign-in happens in your browser.
Only people you invite to your organization. Note that today an invitation is close to full trust — any member can create, deploy and delete anything. To let someone merely view an app, invite them to the app instead. See Work with teammates.
In Europe — your app and its data sit in EU data centres, under European rules. That matters more than usual once your app starts holding other people's details. Our Privacy Policy names every provider involved and says which of them can be reached from outside the EU.
A link. Our Data Processing Agreement is the GDPR Article 28 agreement covering personal data inside what you deploy, and it is already in force — it came with the terms you accepted, it is the same for every customer, and nobody has to sign anything. It names where the data sits, the two sub-processors involved, what happens on a breach and what happens when you leave. Send the URL; it has a version and a date on it.
Money and leaving
Static sites are free in useful quantity, forever, with no card. Anything that runs code needs the paid plan — there are no apps on the free one — and is then a small monthly amount per app, added one at a time. Current numbers: pricing. How the counting works: Plans & what counts.
No — one site. Sites are counted when created, not per page, and adding pages later is free.
Yes. Your files are yours and you can ask your agent for them at any time. Nothing here is a format only we can read: what runs is your project, built into an ordinary container image.
It's gone — no bin and no restore. Deleting a project takes every app inside it. See Take something down.
Still stuck? Fix a broken deploy covers the technical failures, and the toolbelt documents every tool your agent has.