Tau Voice
Team led by TauCAD Founder Richard Fontein and a SmartTrade Fullstack Engineer, experts in open-source AI CAD, TypeScript, and enterprise tooling.
Project Description
Tau Voice: Voice-Enabled Conversational CAD Design
What It Does
Tau Voice adds hands-free voice interaction to Tau, a browser-based CAD platform. Users speak design requests and receive voice responses while watching 3D models generate in real-time. Perfect for designers working with their hands or away from keyboards.
Flow: Voice input (ElevenLabs) → Multi-agent orchestration (LangGraph) → CAD generation → Voice response → Visual preview
How It Works
- Multi-Agent System: LangGraph supervisor routes between specialized agents (research expert, CAD expert)
- Voice Integration: ElevenLabs provides bidirectional speech-to-text and text-to-speech
- Browser Tools: Agents can search the web and browse pages for reference materials
- CAD Generation: Supports multiple kernels (Replicad, OpenSCAD, JSCAD, KCL) for real-time 3D model creation
- Real-Time Updates: Event stream communication keeps UI synchronized with agent actions
Judging Criteria
Working Prototype: Fully functional end-to-end with voice I/O, real-time CAD generation, and error handling.
Technical Complexity & Integration: Multi-agent orchestration (LangGraph), ElevenLabs voice API, browser tool agents (Tavily), multi-kernel CAD support, and robust state management.
Innovation & Creativity: First voice-enabled conversational CAD platform combining hands-free design with sophisticated multi-agent AI.
Real-World Impact: Accessibility for mobility-limited users, hands-on prototyping workflows, democratizes CAD through natural language.
Theme Alignment: Voice + browser + tool agents seamlessly integrated into cohesive autonomous assistant.
Tech Stack
Voice: ElevenLabs API
AI Orchestration: LangGraph, LangChain
Frontend: React 19, Three.js
Backend: NestJS, PostgreSQL (Drizzle ORM)
CAD Kernels: Replicad, OpenSCAD, JSCAD, KCL
Tools: Tavily (web search), OpenAI/Anthropic (LLMs), Event Streams
Setup & Demo
Prerequisites: Node.js 24+, pnpm, Docker
# 1. Install dependencies
pnpm install
# 2. Start database
pnpm db:up
# 3. Configure environment
cp apps/ui/.env.example apps/ui/.env.local
cp apps/api/.env.example apps/api/.env.local
# Add ElevenLabs API key and other required keys
# 4. Start servers
pnpm dev
Access: UI at http://localhost:5173, API at http://localhost:3000
Demo:
- Open web interface → Enable voice mode in chat
- Speak: “Create a parametric box, 50mm wide, 30mm tall, 20mm deep”
- Watch agent generate CAD model with voice feedback
- Continue conversation: “Make it taller” or “Add rounded corners”
- Model updates in real-time with voice responses
Prior Work
Prior Work Description
This project builds upon Tau, an existing open-source CAD platform. All code outside of the elevenlabs branch represents prior work.
Existing Platform
The Tau platform includes:
- Basic frontend with React and code editor
- Backend API with database
- CAD kernel integrations (Replicad, OpenSCAD, JSCAD, KCL)
- Text-based chat interface with AI agents
- File management and 3D visualization
The hackathon work adds voice interaction capabilities on top of this existing infrastructure.
Team
Products & Tools
Additional Links
The development branch containing the implementation of Tau Voice
The main Tau framework, used as a basis for the hackathon project