Stage 3. Generate the complete FastAPI bot project from spec.json. Writes main.py, agent.py, database.py, tools/whatsapp.py, prompt.py, requirements.txt, and .env. Wasender-aware throughout. Runs the bot locally to verify it works before deploying.
Stage 5. Add tools to the live bot one at a time — Google Calendar, Gmail, persistent reminders, human handoff, group chat, voice transcription. Each tool is wired, deployed, and verified before moving to the next.
Stage 6 (optional). Choose where conversation history and OAuth tokens live. Default is SQLite on Render's ephemeral disk; production needs Render Disk, Supabase, or Render Postgres.
Forever-stage. Tweak the bot's personality, refresh expired tokens, diagnose "the bot is acting weird", recover from a disconnected WhatsApp, update tools, view recent conversations. The user lives here once the bot is deployed.
Stage 4. Push the generated bot to GitHub, deploy on Render.com, register the public webhook URL with Wasender, and verify a real round-trip message works.
Stage 2. Hebrew Q&A that produces spec.json — the bot's name, tone, audience, in/out scope, knowledge base, and tools wishlist. This spec drives every later stage.
Stage 1. Sign up at Wasender, get a Personal Access Token, create a WhatsApp session, scan the QR with the user's phone, and verify the connection works end-to-end. Saves credentials to .env.