Trans-N-ai/swama

High-performance MLX-based LLM inference engine for macOS with native Swift implementation

What it solves

Swama provides a high-performance, local machine learning runtime for macOS users, eliminating the need for cloud-based AI services. It simplifies the deployment and use of Large Language Models (LLMs), Vision Language Models (VLMs), and audio models (ASR/TTS) directly on Apple Silicon hardware.

How it works

Built on Apple's MLX framework and written in pure Swift, Swama optimizes inference for Apple Silicon. It offers three primary interfaces: a native macOS menu bar app for easy access, a command-line interface (CLI) for model management and inference, and an OpenAI-compatible API server that allows other applications to integrate with local models.

Who it’s for

macOS users with Apple Silicon (M1/M2/M3/M4) who want to run powerful AI models locally for privacy, performance, and offline access, as well as developers who need an OpenAI-compatible local backend.

Highlights

  • Multimodal Capabilities: Supports text, image (VLM), and audio (speech-to-text and text-to-speech) inputs and outputs.
  • OpenAI Compatible API: Implements standard endpoints for chat completions, embeddings, and audio transcriptions.
  • Smart Model Management: Features automatic downloading from HuggingFace and a system of friendly aliases (e.g., qwen3, llama3.2) to simplify model selection.
  • Native macOS Integration: Includes a dedicated menu bar application for a seamless system-level experience.