oMLX
Native macOS server for fast local AI model inference
Score
60
Évaluation Utilisateur
3.0
Support Gratuit/Essai
Non Pris en Charge
Prix de Démarrage
--
Dernière Mise à Jour
sept. 11, 2026
Aperçu
Qu'est-ce que oMLX ?
oMLX is a native macOS inference server for running local AI models on Apple Silicon using Apple's MLX framework. It provides paged SSD KV caching to reduce time-to-first-token for coding agents, continuous batching for higher throughput, and OpenAI- and Anthropic-compatible APIs. oMLX supports LLM, vision-language, embedding, and reranker models, along with tool calling, MCP integration, multi-model serving, and a native menu bar application with a web dashboard.
À quoi sert oMLX ?
Fonctionnalités Principales
- Paged SSD KV caching for faster responses and reduced time-to-first-token
- Continuous batching for concurrent request throughput
- OpenAI-compatible and Anthropic-compatible APIs
- Native macOS menu bar application
- Web dashboard for model management, chat, and real-time metrics
- Multi-model serving for LLM, VLM, embedding, and reranker models
- Tool calling and MCP integration
- Support for MLX-format models from Hugging Face
- LRU model eviction when memory is limited
- Automatic reuse of Hugging Face and LM Studio model caches
Comment utiliser oMLX ?
Download the signed macOS DMG and drag oMLX to the Applications folder, then configure the model directory, start the server, and download or select an MLX-format model. Alternatively, install from source with Python 3.10+ using the provided Git commands and run omlx serve with a model directory. Connect Claude Code, OpenClaw, Cursor, or other compatible clients through the OpenAI or Anthropic API endpoints.
Outils Alternatifs
Aucun outil alternatif trouvé
Avantages & Inconvénients
Aucune Donnée
Cas d'Utilisation
- Run local coding agents with faster responses on Apple Silicon Macs
- Use Claude Code, OpenClaw, or Cursor with a local AI backend
- Serve multiple local language and vision models through compatible APIs
- Build applications using local OpenAI-compatible or Anthropic-compatible inference
- Test tool calling and MCP workflows without relying on cloud inference
- Run private local chat, embedding, reranking, and multimodal workloads
Avis Utilisateurs
3.0
sur 5
3.0
Aucun avis pour le moment
Soyez le premier à évaluer cet outil !