<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> This project provides a proxy server enabling free access to Claude-like models via NVIDIA NIM, OpenRouter, or local LM Studio, bypassing the need for Anthropic API keys. It targets developers and power users seeking cost-effective and flexible LLM integration for terminal, VSCode, or Discord applications. The primary benefit is significant cost savings and the ability to leverage a wide array of models.

How It Works

This project implements a lightweight proxy server that intercepts standard Anthropic API calls from clients like the Claude Code CLI or VSCode extension. It routes these requests to configured external LLM providers (NVIDIA NIM, OpenRouter, LM Studio), which often expose OpenAI-compatible endpoints. The proxy translates between Anthropic's SSE format and the provider's format, while also performing optimizations like intercepting trivial API calls (e.g., quota probes, title generation) to conserve API quota and reduce latency. It also supports parsing Claude-specific tags like <think>.

Quick Start & Requirements

Installation involves cloning the repository, copying .env.example to .env, and configuring API keys or local endpoints within the .env file. The proxy server is started using uv run uvicorn server:app --host 0.0.0.0 --port 8082. Clients (CLI, VSCode) are then configured to point to http://localhost:8082 with a specific auth token (freecc). Prerequisites include uv (Rust-based Python package installer), Python, and potentially fzf for the interactive model picker (claude-pick). API keys for NVIDIA NIM or OpenRouter are required unless using LM Studio locally, which must be running separately. Links to provider setup: NVIDIA NIM, OpenRouter, LM Studio.

Highlighted Details

• Cost-Effective Access: Offers free usage tiers via NVIDIA NIM (40 req/min) and OpenRouter (free models), alongside unlimited local execution with LM Studio, eliminating direct Anthropic API costs.
• Flexible Provider Routing: Supports NVIDIA NIM, OpenRouter, and LM Studio, enabling per-model mapping (Opus, Sonnet, Haiku) and dynamic provider mixing.
• Intelligent Request Optimization: Intercepts and locally handles five categories of non-essential API calls, such as quota probes and title generation, saving API quota and reducing latency.
• Autonomous Agent Capabilities: Includes a Discord/Telegram bot for remote control, session persistence, live progress streaming, and voice note transcription (local Whisper or NVIDIA NIM).
• Interactive Model Selection: Features a claude-pick command (requires fzf) for dynamically choosing models from the configured provider at runtime.

Maintenance & Community

The project follows standard GitHub contribution workflows for bug reports, feature suggestions, and pull requests. While it integrates Discord and Telegram bots for autonomous agent functionality, direct links to community servers or active developer channels are not provided in the README.

Licensing & Compatibility

This project is licensed under the MIT License, permitting commercial use and modification.

Limitations & Caveats

Docker integration is explicitly not supported at this time. Users must manage API keys for external providers or run LM Studio locally. The project's Python 3.14 badge may indicate a future target or a typo, as this version is not yet released.