ClackChat runs a small language model — Bonsai 1.7B, a 1-bit model — entirely inside your browser using WebGPU. Nothing you type is sent to a server: once the model is downloaded, the conversation stays on your machine.
The first start downloads ≈ 291 MB of model weights from the Hugging Face Hub. Your browser caches them, so later starts are quick.
Expect a small, occasionally confused assistant — it is a tiny model running on your GPU, not a datacentre. Requires roughly 2 GB of free memory.
ClackChat needs WebGPU, which this browser or device doesn't offer. Recent Chrome, Edge, Firefox or Safari on a desktop machine should work.
Starting…
Model:
Bonsai 1.7B
by prism-ml, ONNX conversion
by onnx-community · Apache-2.0.
Engine: Transformers.js
by Hugging Face, Apache-2.0 ·
ONNX Runtime
by Microsoft, MIT