CloudWaddie/LMArenaBridge

What it solves

LMArenaBridge provides a way to interact with the models hosted on LM Arena through a standardized API. It solves the problem of accessing these models programmatically by creating an OpenAI-compatible API endpoint, allowing users to integrate LM Arena models into their own applications or interfaces like OpenWebUI.

How it works

The project acts as a bridge that forwards requests from an OpenAI-compatible API to the LM Arena website. It requires an authentication token (arena-auth-prod-v1) extracted from the user's browser cookies to authenticate requests. To bypass security measures, it can use a userscript proxy or an optional AI-powered reCAPTCHA v2 solver that utilizes YOLO vision models to automatically solve image challenges.

Who it’s for

It is designed for developers and AI enthusiasts who want to use LM Arena's model collection within their own LLM frontends or software tools without manually using the web interface.

Highlights

• OpenAI Compatible: Provides a standard API endpoint for easy integration with existing LLM tools.
• Vision Support: Supports sending images to vision-capable models via automatic upload to LMArena's R2 storage.
• Automated CAPTCHA Solving: Includes an optional YOLO-based vision solver to handle reCAPTCHA v2 image challenges.
• Admin Dashboard: Features a built-in portal for managing authentication tokens and API keys.
• OpenWebUI Integration: Specifically documented for use as a backend for the OpenWebUI interface.