Spawn — One-Click AI Deploy Platform

Push code. Claude configures it. Spawn deploys it. The self-healer fixes it.

Architecture

GitHub Push → Webhook → Deploy Queue (BullMQ)
                              ↓
                    Claude Opus 4.6 Analysis
                    (Dockerfile + RAM + config)
                              ↓
                    Fly.io Machines API deploy
                              ↓
                    Health Monitor (every 30s)
                    → Auto-restart → Scale → Claude heal
                              ↓
                    Email / Webhook notification

Setup

1. Prerequisites

• Bun installed
• Fly.io account + flyctl installed and authenticated
• Supabase project
• GitHub App created
• Redis (local: brew install redis && brew services start redis)

2. GitHub App settings

• Permissions: Contents (read), Metadata (read)
• Subscribe to events: Push
• Callback URL: http://localhost:3001/github/callback
• Webhook URL: http://localhost:3001/github/webhook (use ngrok for local dev)

3. Supabase

1. Create a new project
2. Run apps/api/src/db/schema.sql in the SQL editor
3. Run apps/api/src/db/functions.sql in the SQL editor
4. Enable GitHub OAuth in Authentication → Providers

4. Environment variables

cp apps/api/.env.example apps/api/.env
cp apps/web/.env.example apps/web/.env.local
# Fill in all values

5. Install & run

bun install
bun run dev          # starts both API (3001) and web (3000)
bun run worker       # start deploy worker + health checker in separate terminal

Key files

File	Purpose
apps/api/src/agents/analyzer.ts	Claude Opus 4.6 — analyzes repo, generates Dockerfile
apps/api/src/agents/analyzer.ts (diagnoseCrash)	Claude diagnoses crashes, returns fix
apps/api/src/workers/deploy.ts	Full deploy pipeline (clone → analyze → build → deploy)
apps/api/src/workers/healer.ts	Self-healing loop (health checks + auto-scale + Claude)
apps/api/src/services/fly.ts	Fly.io Machines API wrapper
apps/web/app/deploy/page.tsx	One-click deploy UI
apps/web/app/dashboard/page.tsx	App dashboard

Self-healing logic

Health check fails
  → 1st/2nd failure: restart machine
  → 3rd+ failure: send crash logs to Claude Opus 4.6
      → OOM detected? scale_memory (double RAM)
      → Code error? redeploy_with_fix (patch Dockerfile)
      → Transient? restart
      → Unknown? notify_only (email user)
Memory > 85%? → scale_memory proactively (1.5x RAM)