Deploy to Cloudflare

For an operator standing a deployment up on Cloudflare, which is the reference runtime for Cat Factory. The backend runs as a Worker with D1, Durable Objects, and Workflows. The frontend is a Nuxt SPA on Cloudflare Pages, and per-run coding work executes in Cloudflare Containers.

Your deployment project

Cat Factory ships as reusable libraries on npm (plus a runner image on GHCR and Docker Hub), which you assemble into a small deployment project of two thin packages that depend on the published libraries and point at your own Cloudflare resources:

• Backend: re-exports @cat-factory/worker and ships your wrangler.toml (D1 binding, container image tag, secrets, custom domain).
• Frontend: a Nuxt app that extends @cat-factory/app and ships your Pages wrangler.toml.

To scaffold it, copy the deploy/backend and deploy/frontend example directories from the repo, swap their workspace:* dependencies for the published npm versions, and point the config at your resources. From here you can mix in proprietary agent kinds, extra model providers, and seeded pipelines. See Extending a deployment.

Prerequisites

• A Cloudflare account with Worker and D1 database access.
• Your deployment project depending on the published @cat-factory/* packages.
• A GitHub App configured for authentication and repository operations (see Register the GitHub App).
• LLM provider API keys, or use the Cloudflare Workers AI default (no key required; billed under your Cloudflare account's Workers AI pricing).
• wrangler and pnpm installed locally.

1. Deploy the backend

From your backend deployment project, apply database migrations and deploy the Worker. The D1 migrations ship inside @cat-factory/worker (your wrangler.toml's migrations_dir points at node_modules/@cat-factory/worker/migrations), so they travel with the dependency:

# Review and apply D1 migrations (use your own database name)
wrangler d1 migrations list <your-d1-database> --remote
wrangler d1 migrations apply <your-d1-database> --remote

# Deploy the Worker (builds @cat-factory/worker, then wrangler deploy)
pnpm deploy

2. Deploy the frontend

Build the Nuxt SPA pointing at your backend URL, then publish it to Pages:

NUXT_PUBLIC_API_BASE=https://your-api-domain.com pnpm generate
pnpm deploy

Key configuration in wrangler.toml

The Worker is wired up through bindings in wrangler.toml:

For the full list of secrets and environment variables, including the opt-in web search and tracker settings, see Configuration.

Production topology

┌─────────────────────────────────────┐
│ Nuxt SPA (Cloudflare Pages)         │
│ API base → your Worker domain       │
└──────────────┬──────────────────────┘
               │ REST + WebSocket
┌──────────────▼──────────────────────┐
│ Cloudflare Worker                   │
│ ├─ Hono controllers                 │
│ ├─ D1 database binding              │
│ ├─ Durable Objects (event hubs)     │
│ └─ Workflows (run orchestration)    │
└──────────────┬──────────────────────┘
               │ dispatch per-run jobs
┌──────────────▼──────────────────────┐
│ Cloudflare Containers (ephemeral)   │
│ executor-harness → coding agent     │
│ → Git operations & PR creation      │
└─────────────────────────────────────┘

Why this setup

• Durable Workflows checkpoint each run step, so runs survive restarts and retry from failure.
• Durable Objects give each workspace an event hub that pushes live updates over WebSockets, so there's no polling.
• Containers isolate each run's coding work and Git operations.

Next: set your secrets and toggles in Configuration, or run on your own servers instead with Deploy to Node.js.