NOVA Local AI
An experimental assistant that runs the model in your browser. Chat and primitive-scene generation do not require an OpenAI, Grok, Claude, Gemini, Groq, or Pollinations API key.
1. Choose and load a local model
WebGPU and enough device memory are required. The first load downloads model files from a public CDN and can use hundreds of MB or more of data. Browser and Android device support varies.
2. Chat with NOVA
Load a model, then ask an engineering or coding question.
3. Generate a concept scene and send it to CAD engines
Describe a simple object or assembly. NOVA will try to produce a validated list of primitive parts, then you can send it to the configured Cascade Studio/OpenCascade.js worker or request a STEP export from the FreeCAD worker.
CAD actions require the secured worker to be deployed and configured. Generated shapes are concept geometry, not validated engineering. The embedded CAD editor is not live-synchronized.
This is not a cloud-scale ChatGPT replacement. Smaller local models can make mistakes and may be slow or fail on phones without WebGPU or enough memory. CAD worker actions are attempted only when you press the relevant button; this page does not claim they succeeded unless the worker returns success. The AI does not validate dimensions, structural safety, or manufacturing suitability.