VoiceForge

A "Stripe for Voice Agents" SaaS platform. Developers sign up, define an AI voice agent (STT + LLM + TTS providers, system prompt, tools), and get a hosted, scalable voice agent accessible via WebRTC or phone.

Quick Start

# 1. Start infrastructure
docker compose -f infra/docker/docker-compose.yml up -d

# 2. Install dependencies
pnpm install

# 3. Run database migrations
pnpm db:migrate

# 4. Start API and web (concurrently)
pnpm dev

# 5. In another terminal, start the Python worker
cd apps/agent && uv run python worker.py

Project Structure

/apps
  /api          → NestJS control plane (REST API, auth, billing, sessions)
  /web          → React 19 + Vite + Tailwind dashboard
  /agent        → Python 3.12 Pipecat voice agent worker
/packages
  /shared-types → TypeScript interfaces shared across apps
  /db           → Drizzle ORM schema + migrations
/infra
  /docker       → docker-compose.yml for local dev
  /k8s          → Kubernetes manifests

Tech Stack

• API: Node.js 20, NestJS 10, Drizzle ORM, BullMQ, Passport.js
• Web: React 19, Vite, Tailwind CSS v4, TanStack Router, Zustand, Recharts
• Agent: Python 3.12, Pipecat, LiveKit, Silero VAD
• Database: PostgreSQL 16
• Cache/Queue: Redis 7
• Container: Docker, Kubernetes

Commands

pnpm dev                    # Start API + Web concurrently
pnpm build                  # Build all apps
pnpm lint                   # Lint all apps
pnpm format                 # Format all apps
pnpm test                   # Run all tests
pnpm db:generate            # Generate Drizzle migration
pnpm db:migrate             # Apply migrations
pnpm db:studio              # Open Drizzle Studio

Environment Variables

Copy .env.example to .env in both apps/api and apps/agent, then fill in your values.

Documentation

• CLAUDE.md — Development conventions and coding guidelines
• PLAN.md — Full 16-week technical roadmap
• Features.txt — Functional specification