On-device · Gemma · MediaPipe Web

Character Studio

Define a character on the left, then chat with it on the right. The model loads into your browser the first time you send a message, after which everything runs locally with WebGPU.

Character

Quick start

Spin up a full character and open the chat instantly.

Few-shot example dialogue (optional, advanced)

Separate examples with a line containing only ---. Use {{char}}: and {{user}}: prefixes.

My library

Build a character on the left, then press Start chat to load Gemma on-device and begin the conversation.

Setup

Inference runs via Gemma 3 1B + Transformers.js with WebGPU. The ONNX model is fetched from HuggingFace and cached in the browser automatically. Requires Chrome 113+ with WebGPU.