How to deploy an AI-built app

Choose what the platform should operate

A virtual server gives you control but leaves server patching, database operation and recovery setup with you. A managed application platform such as Render handles service infrastructure; Railway lets you compose services with explicit database responsibilities. Vercel is worth evaluating for a frontend-centric workload.

AppHaven is designed to run AI-built business applications with a managed database, private access and production backups around them. The walkthrough below uses AppHaven and a Node.js invoice tracker. The persistence, permissions and release checks apply on other hosts too.

Start with a working app

The example is an invoice tracker built with Node.js. It already has a screen for creating invoices, a database connection read from DATABASE_URL, and an npm start command. The same steps apply to other business apps; substitute their build and start commands.

You need Git, Docker for the local build check, and an AppHaven account. Your app must listen on 0.0.0.0:3000 and use PostgreSQL for this database example. If it currently uses SQLite, changing the connection string is not a conversion: ask your agent to adapt the database layer and migrations first.

Package the app

FROM node:22-bookworm-slim
WORKDIR /app
ENV NODE_ENV=production
COPY package.json package-lock.json ./
RUN npm ci --omit=dev
COPY --chown=node:node . .
USER node
EXPOSE 3000
CMD ["npm", "start"]

node_modules
.git
.env
.env.*
*.log
backups
exports
*.dump

docker build -t invoice-tracker .

Declare the app and its database

services:
  web:
    type: container
    build:
      dockerfile: Dockerfile
    ports:
      - 3000
    env:
      DATABASE_URL: ${service.db.url}
    depends_on:
      - service: db
        condition: healthy

  db:
    type: postgres
    version: "17"

Make the first database setup explicit

Your application needs tables as well as a connection. Use its migration tool to create the schema, and document the command in the repository. For example, an existing npm run migrate script can run before the web process starts. Confirm the script is safe to run again and that a failure stops startup; do not add a script name that your project does not implement.

For this first deployment, use an empty database and synthetic invoices. Existing production data needs a separate import and verification step. The database guide covers data changes and recovery.

Connect and deploy a preview

1. 1

   Commit a setup branch

   Commit the Dockerfile, .dockerignore, manifest and migration changes on a branch such as deploy-setup. Keep it separate from main.

2. 2

   Create the app

   In your AppHaven project, create an application and choose its region. In Source, connect the repository and select main as the production branch.

3. 3

   Add runtime settings

   In Settings, add the application’s base URL and other plain settings as Variables. Add API keys as Secrets. Use sandbox keys for external services while testing; a separate database does not stop a preview from sending a real email or charging a card.

4. 4

   Deploy deploy-setup

   Choose the setup branch in Deploy. Follow the build log, then open the preview URL when the deployment is Running. Keep access private. The console walkthrough shows these controls.

Check the result before releasing

Release it to your team

Merge the reviewed setup into your production branch and deploy that branch. Production has its own database: run the schema migrations there and check the app again. Preview records are test data, not an import into production.

Invite a colleague through the project’s Members page. In Backups, check the production restore window and choose the retention you need. AppHaven handles backup capture; restoration currently goes through support. Keep the app and deployment names in your release notes so you can request help quickly.

If the deployment fails