About this project

Rapid-MLX is a high-performance local AI engine designed specifically for Apple Silicon (M1-M4). It provides a drop-in replacement for OpenAI and Anthropic APIs, allowing existing clients like Claude Code, Cursor, Aider, and LangChain to connect without adapters. Key features include: - **Performance**: Claims up to 3x Ollama's throughput with native MLX kernels, continuous batching, prompt caching, and quantized KV cache. - **API Compatibility**: Supports `/v1/chat/completions`, `/v1/responses`, `/v1/messages`, `/v1/embeddings`, `/v1/audio/*`, `/v1/videos`, and `/v1/images/*`. - **Model Support**: Text, vision, audio (TTS/STT/voice cloning), video generation, image generation, and embeddings. Includes a model catalog with aliases for various sizes and memory requirements. - **Agent Integration**: Wire-verified compatibility with 12 agent CLIs (e.g., Claude Code, Codex CLI, Aider) and 3 Python frameworks (LangChain, PydanticAI, smolagents). Tier-1 agents are re-verified end-to-end on real weights before each release. - **Installation**: Available via Homebrew, pip, or a guided installer. Desktop app for macOS. - **Use Cases**: Terminal chat, OpenAI-compatible server for apps, agent backends, and benchmarking. - **Extras**: Optional packages for image, video, audio, and system-one (typed decisions) capabilities. Rapid-MLX is positioned as a fast, local alternative for developers and AI enthusiasts on Apple hardware, emphasizing performance, compatibility, and ease of use.