Deploy an app built by an AI coding agent
Using Claude, ChatGPT, Cursor or another AI agent to build your app? Here is what to ask for so the code runs on Schokr first time.
Updated · 2 min read
AI coding agents can write an app that runs on Schokr, but they cannot deploy it for you: you upload the code with the file manager and set the startup settings yourself. The trick is to tell the agent about Schokr's environment up front so the code it produces fits.
What the agent needs to know
Paste this into your prompt, adjusting the runtime:
This app will run on a hosted Linux container (Schokr) using the Node.js template.
Requirements:
- Listen on host 0.0.0.0 and read the port from an environment variable APP_PORT.
- Load configuration and secrets from a .env file. Never hard-code tokens.
- List every dependency in package.json, with a "start" script.
- No Docker, no docker-compose, no systemd, no nginx config.
- There is no database server. Use SQLite in a file inside the app folder,
or a DATABASE_URL for an external provider.
- Keep memory use under 512 MB.
- Log to stdout/stderr only.
- Tell me the exact command to start the app and which file is the entry point.
For Python, swap in requirements.txt and the server you want (gunicorn, uvicorn, or a plain script).
What you should get back
Check the agent's output contains:
| Item | Example |
|---|---|
| Entry file | index.js, main.py |
| Dependency file | package.json, requirements.txt |
| Start command | node index.js, uvicorn main:app --host 0.0.0.0 --port 25571 |
| Env var list | APP_PORT, BOT_TOKEN |
.env.example | With empty values |
Deploy steps
- Create a server from the matching template (Node.js, Bun or Python) in the console.
- Note the port in the Network section and put it in
.envasAPP_PORT. - Upload the files with the file manager. Do not upload
node_modulesor any venv the agent created locally. - Set the main file or startup command in your server's Startup settings.
- Start the server and read the console output.
When it fails, give the agent the logs
Copy the full error from the console and paste it back to the agent with: "This error appears on startup on the hosted server. Fix it without changing the host, port or environment rules above." Most first-run failures are one of:
- Listening on
localhostor a fixed port. See binding to 0.0.0.0. - A missing package in the dependency file. See module not found.
- Using a feature that needs a newer runtime version than the server has.
- Expecting Docker, a Redis server or a local Postgres that does not exist.
Things agents often get wrong
- Writing files outside the app folder. Keep data paths relative to the app.
- Hard-coding secrets pulled from your chat. Move them to
.envand rotate any token you pasted into a chat. - Dev servers in production, such as
flask run,next devor--reload. Ask for the production command. - Assuming HTTPS. Schokr serves your app on plain HTTP at your server's address. See using your own domain.
Review before you run
Read through what the agent wrote, especially anything that deletes files, sends requests to unknown URLs or handles user data. You are responsible for what runs on your server.