How to deploy code an AI wrote for you

Why this step is harder than the code

The AI wrote the app, but deploying it the traditional way means choices it cannot make for you: a host, a database, an auth provider, HTTPS certificates, environment variables, a build pipeline. That stack is exactly what Benmore collapses: apps run on a platform where hosting, the database, auth, and the API are built in, so "deploy" stops being a project and becomes a step.

Path 1: Keep working in your external AI client

Connect Claude, ChatGPT, Cursor, or Codex to Benmore over MCP, provide the code in that client, and say what it is: "This is a habit tracker. Make it a real app I can log into from my phone." The client adapts the code and deploys it through Benmore's tools while you keep iterating in the same conversation. If the code is a single HTML file, see deploying an HTML page.

Check what the generated code needs

Benmore runs HTML/TSX with Prisma and YAML. A Node server, Next.js server routes, Python backend, or arbitrary container cannot be uploaded unchanged. Build static frontends in their original project and copy the browser output, or adapt backend behavior to Benmore APIs and flows. Review authentication and data access before importing real data.

Path 2: Keep your AI, add a deploy target

If you work in Codex, Claude Code, Cursor, or another coding agent, give it Benmore as the deploy target instead of changing tools:

brew install Benmore-Studio/benmore/benmore-cli
benmore login
benmore bootstrap   # installs shared Codex/Claude skills and sets up your workspace

On Linux or CI, replace the Homebrew line with curl -fsSL https://benmore.ai/install-cli.sh | sh.

From then on installed Claude Code hooks ship the agent's edits to a live hosted dev URL; Codex, Cursor, and other agents run benmore push <file> explicitly. The agent verifies its own work in a real browser and ships to production when you say go. The "works on my machine" stage disappears because there is no your-machine stage: the dev environment is the live URL.

What you get either way

• A live HTTPS URL on the first push, with a custom domain when ready
• Auth, a per-app database, and a typed REST API without configuring any of them
• Hot reload in about a second per change
• Per-app source history; source reverts do not undo data migrations or external effects

Verify before sharing

Use the agent workflow guide to inspect live contracts, push files, exercise the real browser journey, and promote source. To run the app on your own infrastructure instead, follow self-hosting.

Frequently asked questions

ChatGPT gave me code. How do I make it a real website?

Connect that AI client to Benmore over MCP and describe what the code should become, or put it in an app directory and run benmore deploy.

Does this replace Claude Code or Cursor?

No, it is the deploy target for them. The Benmore CLI installs an agent skill so your coding agent builds directly against a live hosted environment.

What languages and frameworks does it run?

Benmore apps are TypeScript pages over a Prisma schema with YAML for server logic. Your external AI client can adapt HTML/JS/React-style code into that shape; Benmore is not a general container host for arbitrary backends.

What does hosting cost?

Plans are priced by app capacity and publishing, with a free tier to start, not per prompt or per seat. See pricing.