kitio.aiAdded Nov 5, 2026
oMLX
FreeNative macOS LLM server with SSD KV caching for coding agents
What is oMLX?
oMLX is an open-source inference server for Apple Silicon Macs, built on MLX, that serves local models through OpenAI- and Anthropic-compatible APIs. It persists the KV cache to SSD so coding agents such as Claude Code, Cursor and OpenClaw get fast responses on repeated prompts.
Coding agents reuse long prompt prefixes, and oMLX stores cache blocks on disk in safetensors format so those prefixes are restored instead of recomputed, even after a restart. Continuous batching handles parallel requests, and LLM, vision, embedding and reranker models can be loaded at the same time.
A native menu bar app starts and monitors the server, and a web dashboard handles model downloads, chat, metrics and the config command for each client tool. It supports common tool calling formats and MCP, is released under Apache 2.0 and requires macOS 15 or later.
Features
- Paged SSD KV cache
- OpenAI and Anthropic APIs
- Continuous batching
- Multi-model serving
- Menu bar app
- Tool calling and MCP
Tags
Releases
+ Submit new release- Listed
oMLX was added to the directory.
Organization
oMLX
Listed by kitio.ai · unverified owner
Pricing
- Pricing model
- Free
- Paid options from
- 100% free
Sponsored
Your AI tool here
A Featured launch puts your tool in this spot and on the home page for a week.
Top oMLX alternatives
View allModels & APIs
Desktop app for chatting with open-source LLMs fully offline
Models & APIs
Apple's array framework for running and training models on Apple Silicon
Models & APIs
Open-source Mac app for chatting with local LLaMA models
If you liked oMLX
Free native app to run AI models offline on your computer
FreeModels & APIs6 views2 clicks to the websiteThe main hub for open models, datasets and AI demo apps
Freemium · $9/moModels & APIs14 views6 clicks to the websiteMade in United StatesSearch engine and vector database for AI apps and agents
Free trialModels & APIsMade in NetherlandsRun open models locally with one command, plus optional cloud models
Freemium · $20/moModels & APIs3 views4 clicks to the websiteMade in United States
Pros and cons
Pros
- Fast repeat turns for coding agents
- Drop-in API for common tools
- Free and open source
Cons
- Only runs on Apple Silicon Macs
- Requires macOS 15 or later
- Large models need lots of memory
Prompts & results
No prompts yet. Add your own prompt and output to help others use this tool.
Share a prompt
Reviews
Q&A
What is oMLX?
oMLX is an open-source inference server for Apple Silicon Macs, built on MLX, that serves local models through OpenAI- and Anthropic-compatible APIs. It persists the KV cache to SSD so coding agents such as Claude Code, Cursor and OpenClaw get fast responses on repeated prompts.
Does oMLX work with Claude Code?
Yes. It exposes an Anthropic-compatible /v1/messages endpoint and an OpenAI-compatible API, and the dashboard generates the config command for each tool.
How much does oMLX cost?
oMLX is 100% free.
Is there a free version of oMLX?
Yes. oMLX is completely free to use.
What are the main features of oMLX?
Paged SSD KV cache, OpenAI and Anthropic APIs, Continuous batching, Multi-model serving, Menu bar app, Tool calling and MCP.
Who makes oMLX?
oMLX is made by oMLX.
What are the best alternatives to oMLX?
Popular alternatives include GPT4All, MLX, LlamaChat.