Raw LLM – Local Proxy UI with Multi-Key Rotation and Unfiltered Access

[latent-unshackled]

Hi all,

I built Raw LLM—a lightweight, open-source React/Express frontend and local proxy designed for raw model testing and security research. It bypasses standard web frontend filtering by talking directly to API endpoints.

• 📦 Repo: https://github.com/latent-unshackled/raw-llm
• 🌐 Demo: https://raw-llm.onrender.com/

Core Features

• Zero Proxy Filtering: Direct provider access supporting system overrides, preset injections, and smart prefills.
• Key Rotation: Automatic failover to backup API keys when rate-limited.
• Compatibility: Works out-of-the-box with Mistral, Anthropic, OpenRouter, Together AI, or any custom OpenAI-compatible endpoint (Ollama, LocalAI).
• Local Privacy: All API keys and chat history stay in your browser's local storage.

How it Works (server.ts)

The Express proxy intercepts requests, processes your research-mode configurations (like layered payloads and system overrides), and streams responses straight back to the React UI in real-time.

Feel free to check out the code, run it locally, or open an issue if you have feedback!