Download
MLX-Serve is a native AI inference server and desktop application for running open models locally on Apple Silicon Macs. Its Zig-based backend and Swift interface operate without requiring Python or a cloud account. It exposes OpenAI, Anthropic, and Ollama-compatible APIs for connecting development tools and AI applications. Users can download and manage models through a graphical interface, command-line tools, or a web console. Supported workflows include chat, coding assistance, image generation, video, music, speech, and 3D content generation. Speech transcription and voice cloning extend its multimedia capabilities. Models load on demand, and supported configurations can distribute inference across multiple Macs. The platform also supports sandboxed agent tools, document conversations, and local automation.
Features
- Local AI model inference optimized for Apple Silicon without Python dependencies
- OpenAI, Anthropic, and Ollama-compatible APIs for external application integration
- Native desktop interface with chat, model downloading, and management tools
- Image, video, music, speech, and 3D generation with compatible models
- Speech transcription, voice cloning, and local coding agent integration
- On-demand model loading with multi-Mac distributed inference and agent tools
Project Samples
