Your AI wrote the app. Erawan puts it online.

What it is

Erawan is a hosting platform whose primary user is the agent, not a person with a dashboard open. Claude Code, Cursor, Windsurf, Codex, Gemini CLI, Copilot — whichever one is writing your code — connects to Erawan and deploys through it. What comes back is a link on your-app.erawan.app that anyone can open, over HTTPS, with a certificate already in place.

That inversion is the whole product. Every other platform assumes a developer will read a dashboard, write a config file, and click Deploy. If you have never heard of DNS, a container or a reverse proxy, none of that is a small step — it is the step. So the deploy loop here is made of tools an agent calls and reads: it deploys, it reads the build log, it fixes its own mistake, it deploys again. You watch a link appear.

What a deploy actually does

Not a black box. When your agent deploys, in this order:

1. It is read before it is built. A static check looks for the things that cost real money when they ship — an AWS or Stripe key committed by accident, a .env baked into the image, debug mode left on, a wide-open CORS rule with credentials. Anything certainly wrong blocks the deploy and comes back as something the agent can fix. Anything merely suspect is a warning, because a check that refuses working code teaches people to ignore it.
2. It works out what your project is. A Dockerfile if you wrote one; otherwise Python, or a frontend build, or Node, or plain files. A React or Vite site is built once and then served as files — no container, no server process, nothing metered.
3. It builds, in our cluster. Never on your machine, and never on the machine that holds anyone's account records.
4. It keeps the version. Every successful deploy is a git commit and a tag in a repository the platform keeps for you. Version 7 does not erase version 6 — roll back and the previous image is still there to run.
5. It answers with a URL, and with logs the agent can read if it did not come up.

Your first deploy needs no account

No sign-up form, no credit card, no email confirmation before you have seen whether the thing works. Your agent asks for a guest session, deploys, and hands you a live link. That link stays live for 24 hours. Sign up free inside that window and the apps you already made come with you — the URL you already sent to someone keeps working, because signing up fills in the account on the same records rather than moving them somewhere new.

We built it this way because the alternative is asking for an email address in exchange for a promise. This way you get the thing first.

Start from wherever you are

Four routes, ordered by how much you have to install. The first needs nothing at all — not a download, not an account, not a card — and it is the honest place to begin if the sentence "your coding agent" has not meant anything to you so far.

0. Nothing installed — use the lab on our front page

Describe what you want in ordinary words. An agent builds it in a sandbox we run, you watch it work, and it ends at an address you can open on your phone. The Website tab runs signed out; the tabs that need a database or a backend ask you to make an account first, which is free and takes an email address.

Open the lab on the front page →

1. You already chat with an AI — paste one line

If you have a chat open with something that can read a web page, that is enough. Paste this and it fetches our playbook itself; there is nothing for you to learn or configure.

2. You have the Claude app — connect it once

In its settings, where connectors or MCP servers are added, paste the address below. Connecting opens a browser once to approve it — and approving needs no account either. After that, asking Claude to put something online is a normal sentence in a normal conversation.

3. You want it on your own machine

The route that keeps working when the project outgrows a chat window: the agent can see your files, run things, and read what went wrong.

1. Install an editor that has an agent in it — VS Code with Claude Code added, or Cursor, which comes with one. Either is a normal application: download, open, done.
2. Make a folder for your project, open it in the editor, and start the agent.
3. Have it run the line below, once. It installs the erawan command and the playbook that tells an agent how to use it.

On Windows the same thing is irm https://erawan.cloud/install.ps1 | iex.

You do not have to pick the last one to be doing it properly. Plenty of real, running apps were started by somebody who opened the lab, liked what came out, and only installed anything at all a week later.

What an app gets

The parts you would otherwise be shopping for separately, provisioned by the same agent that wrote the code, with the connection details injected into the app for you — you never copy a password from one screen into another.

• A Postgres database, made on request. On the smaller plans it is a database of your own on a shared instance; on Pro and Max it is an instance of your own.
• Redis and S3-compatible storage, the same way, from Hobby up.
• Email that sends without you owning a mail provider — your app posts a message, we deliver it, and the provider key stays ours.
• A disk for uploads, generated files and caches.
• Scheduled tasks and one-off jobs, running your app's own image — a nightly report, a database migration.
• Logs, error rates and traffic per app, with no setup and nothing to instrument, readable by you in the console and by your agent as data it can act on.
• A custom domain with automatic TLS, on the paid plans.

What it costs

Free here is not a trial with a clock on it. It is a small plan that stays free, and it runs a real process — not just files.

Free฿0 forever3 websites and 1 always-on service (256MB), 100MB Postgres, 100MB disk, HTTPS, version history.

Hobby฿99 / month2 services (1GB total) and 20 sites, Redis and file storage, your own domain.

Pro฿599 / month10 services (2GB total) and 50 sites, a Postgres instance of your own, rewindable 7 days back.

Max฿1,290 / month20 services (4GB total), unlimited websites, 2GB of your own Postgres, rewindable 30 days back.

A flat price, never metered by request or by gigabyte transferred, so there is no bill that arrives as a surprise. What you pay for is running more than one thing at once.

Where it runs, and what we don't claim

Customer apps and their databases are served from a dedicated server in Thailand, so the round trip for a Thai visitor is a domestic one. Accounts, billing records and the control plane are in Singapore, in an enterprise data centre — separate on purpose, because the machine that runs strangers' code should not be the machine that holds everyone's records.

Every app runs in its own namespace with its own network policy, and app configuration — database URLs, tokens, API keys — is encrypted at rest under a key belonging to your account alone. Disks are encrypted; what is on a drive removed from either building cannot be read.

Backups, stated exactly, because this is where hosting pages are vaguest. Your code, the platform's own records and your app's disk are copied offsite every night, encrypted before they leave. On Free and Hobby, each database on the shared instance is dumped separately every hour and those dumps are included in the offsite copy. Hobby can rewind to one of those hourly points for seven days.

On Pro and Max — the plans where the database is an instance of your own — the database is archived continuously and can be rewound to any minute inside the window, seven days on Pro and thirty on Max. That archive is included in the offsite copy too.

Start

If you have never done any of this: open the lab on our front page and type what you want. Nothing to install, no account to start, and the thing it makes is real — a live address, not a picture of one. It stays for 24 hours unless you sign up free to keep the app and its address.

If you already have a coding agent, paste this into the chat you have open. You need no account to start this way either, and you will have a link before deciding whether to sign up and keep it.