How it works

App-Agent is a fork-based platform. When you sign up, you're not logging into our SaaS — you're forking a GitHub repo we maintain into your own GitHub organization. From that moment on, the code is yours. Ours does not touch your production.

The four pieces

1. Your fork of core. A Nuxt 4 monorepo that holds a marketing site, a control plane, a set of installable templates, and shared components. You edit it like any repo. It deploys to your own Cloudflare Pages account.

2. Templates. Ready-to-install app skeletons — a marketing landing, an admin dashboard, a customer portal, an AI chat interface, and more. Each is a working Nuxt app that plugs into your core fork. You install a template by pasting a one-line prompt into Claude Code, Cursor, or Codex, and the AI does the six install steps for you.

3. The AI agent. Claude (via Claude Code, Cursor, or Codex — you pick) has full read/write access to your codebase during Pro-tier sessions. It reads your existing code, understands the layer cascade (see below), and ships features. On the Free tier, the same agent works in read-only mode.

4. Us. The small team behind App-Agent runs weekly sessions with Pro customers — reviewing what shipped, planning what's next, unblocking things the AI can't figure out alone. Enterprise gets dedicated support and BYO-API-key.

The layer cascade

Your fork is structured as three layers:

core/           ← the platform. We maintain this. You pull updates.
organization/  ← your brand. Name, colors, logo, defaults.
apps/          ← your actual apps. One folder per app.

Nuxt merges these three layers automatically at build time. Change your brand color in organization/app.config.ts → every app updates. Pull a new template into apps/support-portal/ → it inherits core's components and organization's brand without you writing a single import.

This is why the AI agent can be trusted to touch your code: the interfaces between layers are small and explicit. It's hard for the agent to break core from inside an app, because apps only extend core; they don't rewrite it.

What a typical week looks like

1. Monday: you tell us (or the AI, in a shared thread) what you want built this week
2. Tuesday–Thursday: the AI ships PRs against your fork. You review + merge (or ask for changes).
3. Friday: 60-minute session with us to review what landed and plan the next batch

Repeat. On average, our customers ship 2–5 meaningful changes per week — features, bug fixes, brand refreshes, new pages.

Ownership at every step

• Code lives in your GitHub org
• Deployments run on your Cloudflare account (or AWS, or GCP — we support all three)
• Data lives wherever you host — never on App-Agent servers
• If you cancel, you keep everything and the AI agent access is the only thing that goes away

There is no vendor lock-in because there is no vendor infra. App-Agent is a codebase + a team, not a hosted product.

What comes next

• What you get breaks down the concrete deliverables.
• Templates explains which template to pick when.
• Ownership, hosting & security covers the compliance questions.