GerbilGerbil
Browser Runtime

Playground

Run Gerbil directly in your browser. No installation required.

Requires WebGPU support (Chrome 113+, Edge 113+)

Loading playground...

How it works

  • 01Models are downloaded to your browser's cache on first use (one-time download)
  • 02Inference runs locally using WebGPU for GPU acceleration
  • 03Your prompts never leave your device. Completely private

Chat Models

  • qwen3.5-0.8b · 404MB
  • lfm2.5-350m · 199MB
  • lfm2.5-2.6b — agentic, 1.5GB (desktop/iPad)
  • qwen3.5-4b — 2.5GB (desktop)
  • Text + vision, 262K context
  • Thinking mode (CoT)

Vision

  • qwen3.5-0.8b · 404MB
  • Image understanding
  • OCR & analysis

Text-to-Speech

  • kani-tts-2 · on-device
  • Native codec-LM synthesis
  • 22.05 kHz mono PCM

Speech-to-Text

  • moonshine-base · encoder-decoder ASR
  • Native raw-waveform transcription
  • English, on-device

Tips

  • ·Use Cmd+Enter to generate/speak
  • ·Enable thinking mode for math and reasoning
  • ·First use downloads models (cached after)
  • ·All processing happens on-device

// Want to use Gerbil in Node.js? Check out the docs