Inferbox

Run AI models directly in your browser — no server, no API key required. Load a model and start chatting.

Click here to load a model

Enter to send Shift+Enter new line

Model Settings

Model Format

Format / Loading Method

Loads ONNX model from Hugging Face Hub via Transformers.js. Supports WebGPU acceleration.

Model Source

HuggingFace Repo ID

Repo must have ONNX files. Browse: huggingface.co/models?library=transformers.js

Execution Backend

WebGPU

WASM/CPU

WebGL

WebGPU uses GPU for fast inference. Falls back to WASM/CPU if unavailable. WebGL uses WebGL compute (fallback to WASM).

Quantization / Dtype

Model dtype

q4 recommended for browser. Applies to ONNX models only.

Multimodal (mmproj)

mmproj URL (optional)

CLIP/vision projection model for multimodal inference (LLaVA etc.). When set, image & audio uploads are enabled.

Generation

Max new tokens

512

Temperature

0.7

Top-P

0.9

System Prompt

Downloading