LocalGateAI turns your Mac into a private AI endpoint. Point Cursor, Cherry Studio, or any tool that supports a custom base URL at your own machine — and run on local models instead of cloud APIs.
Private by default · Zero API cost · Apple Silicon native
The AI tools you use every day — coding assistants, chat clients, agent builders — all talk to cloud APIs. That means usage fees, rate limits, and your prompts leaving your machine. LocalGateAI sits quietly on your Mac and speaks the same language those tools already understand. Switch one setting — the base URL — and everything keeps working, except now it runs on a model on your own hardware.
Choose from built-in models tuned for Apple Silicon and click activate. First time downloads automatically; after that it's offline.
LocalGateAI shows you a Base URL and API Key. One click to copy.
Drop them into Cursor, Cherry Studio, Dify, or anything with a custom endpoint field. Done — your tool now runs on the local model.
Anything that lets you set a custom base URL — Cursor, Cherry Studio, Dify, Coze, Continue, and more. No plugins, no SDK.
Drop-in replacement for the OpenAI chat API. Real token-by-token streaming, so your tools feel responsive.
The gateway binds to localhost only. Your prompts and responses never touch the network. No telemetry, no accounts, no logs.
Built on Apple's MLX framework — runs natively on M-series chips. No Python, no containers, no Ollama to install.
Tuned for the models that actually fit on a laptop. Runs comfortably on 16GB Macs.
Follows your system language, or switch in one click.
| Model | Best for | Size |
|---|---|---|
| Llama 3.2 3B | Everyday chat & agents | ~2 GB |
| Qwen2.5 1.5B | When memory is tight | ~1 GB |
| Qwen2.5 7B | Complex reasoning, long text | ~5 GB |
| Qwen2.5-Coder 7B | Coding, debugging, completion | ~5 GB |
More models over time. All run fully offline once downloaded.
Local-first isn't a feature here. It's the whole point.
If you can paste a URL into a settings field, you're set. The hard part — running the model — is handled for you.
Any tool with a "custom base URL" or "OpenAI-compatible endpoint" option. That includes most modern AI coding and chat apps.
The app is the only thing you get. There are no API fees, no subscriptions, no per-token charges — the model runs on your own hardware.
A Mac with Apple Silicon (M1 or later), macOS 14+. 16GB of memory recommended for the 7B models; 8GB is fine for the smaller ones.
Yes. The gateway is localhost-only and there's no telemetry. Your prompts never leave the machine.
Built-in models cover most needs at launch. Custom model support is on the roadmap.
LocalGateAI is available now on the App Store. Free to download.
Apple Silicon · macOS 14+