Omlx

oMLX is a native macOS inference server built on Apple's MLX framework, designed to run local language models on Apple Silicon Macs. It targets the slow first response that coding agents suffer when a local model has to reprocess a long prompt, and it does so with a paged key value cache that is persisted to SSD, so previously seen prompt prefixes are restored from disk rather than recomputed, including across server restarts. The server exposes an OpenAI compatible API and a native Anthropic style messages endpoint, which lets tools such as Claude Code, OpenClaw and Cursor point at a local model. It supports continuous batching for concurrent requests, serves language, vision language, embedding and reranker models at the same time with least recently used eviction when memory runs low, and handles tool calling in several formats along with MCP tool integration. A menu bar app starts, stops and monitors the server, and a web dashboard covers model downloads, chat and live metrics. The vendor describes the app as signed, notarized and not built on Electron. The project is released under the Apache 2.0 license and is distributed as a DMG download with source on GitHub. It requires macOS 15 or later on Apple Silicon. The site publishes benchmark tables measured on an M3 Ultra with 512 GB of memory. No paid plans or pricing page are published.