AI-CAD

Open-source, AI-driven parametric CAD platform. Describe designs in plain English, generate 3D models, and export to manufacturing formats.

Features

• NLP-to-CAD — Chat with an LLM to generate parametric 3D models from natural language
• Feature-Based Parametric Modeling — Add/remove/toggle features that drive Replicad code generation: extrude, revolve, fillet, chamfer, boolean, mirror, pattern
• Real-time 3D Viewport — Three.js/React Three Fiber with orbit controls, wireframe, edge highlighting
• Code Editor — Monaco editor for Replicad TypeScript with live execution
• Engine Console — Real-time logging of WASM engine events, execution timing, and mesh statistics
• Runtime LLM Configuration — Switch between OpenAI / Anthropic / Ollama providers and models from the Settings page
• Format Conversion — STEP, IGES, STL, OBJ, FBX, GLTF, USD, 3MF, DXF, BREP (18+ formats)
• Assembly Management — Mate definitions: coincident, concentric, distance, angle, tangent, gear
• Universal Parts Consciousness — Intelligent parts library tracking part relationships, failure modes, and usage patterns (UPC)
• MCP Server — Expose CAD operations as Model Context Protocol tools for AI agents
• Export Targets — Unity, Unreal Engine, Web, 3D Print presets

Tech Stack

| Layer | Technology | |-------|-----------| | Framework | React 19, TypeScript, Vite 6 | | 3D Engine | Three.js, @react-three/fiber 9, @react-three/drei 10 | | CAD Kernel | Replicad (OpenCASCADE WASM) | | State | Zustand (with persist middleware for settings) | | Routing | TanStack Router (file-based) | | Data | TanStack React Query | | Editor | Monaco Editor | | LLM | OpenAI / Anthropic Claude / Ollama (runtime-configurable) | | Auth | Azure AD (MSAL) — optional, install separately | | Styling | Tailwind CSS |

Quick Start

# Clone
git clone https://github.com/vespo92/AI-CAD.git
cd AI-CAD

# Install dependencies (Node 22+ required)
npm install

# Copy environment config
cp .env.example .env

# Start dev server
npm run dev

Open http://localhost:5176

The app runs in offline mode by default — you can write Replicad code in the editor and see 3D models without any backend services. To enable AI chat features, either:

• Set VITE_LLM_API_KEY in .env, or
• Configure your LLM provider at runtime via Settings (gear icon in toolbar)

Architecture

src/
  components/
    viewport/       # 3D rendering (Three.js canvas)
    chat/           # LLM chat interface
    editor/         # Monaco code editor
    feature-tree/   # Parametric feature tree (drives code generation)
    parts/          # Parts library browser
    assembly/       # Assembly management
    convert/        # Format conversion
    sketch/         # 2D sketch editor
  lib/
    cad-engine/     # Web Worker CAD execution + feature codegen
    api/            # HTTP clients (LLM, CAD service, UPC)
    mcp/            # Model Context Protocol server
    store/          # Zustand state management (cad, console, llm)
    auth/           # Azure AD (optional stub)
  routes/           # TanStack file-based routes
  types/            # TypeScript type definitions

CAD Engine

The CAD kernel runs in a Web Worker using Replicad (a TypeScript wrapper around OpenCASCADE compiled to WASM). This keeps the UI responsive during expensive geometry operations.

// Example: Create a box with a fillet
const shape = drawRoundedRectangle(40, 30, 5)
  .sketchOnPlane("XY")
  .extrude(15)
  .fillet(2, (e) => e.inDirection("Z"));

return shape;

Feature-Based Modeling

Click the + button in the Feature Tree panel to add parametric features. Each feature maps to generated Replicad code. Toggling visibility or deleting features automatically regenerates and re-executes the model.

Supported features: Sketch, Extrude, Revolve, Fillet, Chamfer, Shell, Boolean Cut/Fuse, Mirror, Pattern.

MCP Server

AI-CAD exposes its capabilities as an MCP server, allowing any AI agent to create and manipulate CAD models:

# Run standalone MCP server
npm run mcp:server

Available tools: execute_code, create_primitive, modify_shape, export_model, import_model, describe_model, send_to_external

Docker

# Build
docker build -t ai-cad .

# Run
docker run -p 8080:80 ai-cad

For production with backend services, pass build args:

docker build \
  --build-arg VITE_LLM_GATEWAY_URL=https://your-llm-gateway.example.com \
  --build-arg VITE_CAD_SERVICE_URL=https://your-cad-service.example.com \
  -t ai-cad .

Backend Services (Optional)

AI-CAD's core modeling works standalone. For the full pipeline, you can run these companion services:

| Service | Purpose | Default Port | |---------|---------|-------------| | LLM Gateway | Proxies LLM requests (OpenAI/Anthropic/Ollama) | 8092 | | CAD Service | Format conversion engine (FreeCAD/OpenSCAD) | 8090 | | File Proxy | File upload/download storage | 8091 | | UPC | Universal Parts Consciousness database | 8093 |

License

MIT