How it works
Local LLM Chat
Messages go to a local worker running Qwen3.5 2B. The conversation stays in memory on your device, and the UI updates as tokens are generated.
- Uses a 4-bit model with WebGPU in a dedicated worker
- Downloads about 1.6 GB on first use, cached and shared with Number Extractor
- Streaming updates keep the UI responsive
- Supports text conversations; answers can be wrong, especially arithmetic
Model: Qwen3.5 2B ONNX, based on Qwen3.5 2B under the Apache 2.0 license.
graph TD
A[User input] --> B[Worker prompt]
B --> C[Qwen3.5 2B model]
C --> D[Token stream]
D --> E[Chat UI]