Skip to main content

ccr chrome-bridge

Start the Chrome On-Device model bridge for Gemini Nano.

ccr chrome-bridge

Description​

Starts a bridge process that connects to Chrome's Prompt API via the Chrome DevTools Protocol (CDP), allowing the router to use Chrome's built-in Gemini Nano model (~4GB local LLM).

Must be run on the host machine (not inside Docker), as it needs direct access to Chrome's debugging port.

How It Works​

  1. The bridge connects to Chrome's remote debugging port (--remote-debugging-port=9229)
  2. It communicates with the Prompt API to send prompts and receive responses
  3. The router's chrome-on-device transformer translates between OpenAI Chat Completions format and the Prompt API
  4. Responses are streamed back through the router as standard SSE

Features​

  • Zero API cost — Fully local inference on your device
  • Streaming support — Full SSE streaming
  • Structured output — Uses responseConstraint for reliable JSON tool calls
  • Automatic stall recovery — Retries with higher temperature if the model stalls

Prerequisites​

  • Google Chrome (Canary or Dev) with Gemini Nano enabled
  • Node.js installed on the host

For detailed setup instructions, see the Chrome On-Device Integration Guide.