Install rapid-mlx
One MLX engine, two ways to run it — a one-line terminal install or the desktop app. Everything runs on your Mac: no accounts, no cloud, no per-request billing.
Two ways to run it
Same engine, two packagings — pick the one that fits how you work.
Desktop app GUI
Click, pick a model, start chatting. No terminal, no config — signed and notarised for macOS.
Download for Mac What's inside the app →Terminal & server CLI
An OpenAI-compatible server on localhost — drop-in for Cursor, Claude Code, Aider, and anything that speaks the OpenAI API.
Requirements
- Chip
- Apple Silicon — M1 or newer
- macOS
- 14 (Sonoma) or newer
- Python
- 3.10+ — auto-installed if missing
- Disk
- Engine is tiny; models download on first serve
- Memory
- Pick a model sized to your RAM — live picker →
Terminal & server install
One command. It probes for Python, sets up an isolated environment, and puts
rapid-mlx on your PATH — no sudo, nothing to sign up for.
$ curl -fsSL https://rapidmlx.com/install.sh | bash
What this command does
- Probes for Python 3.10+ and installs a self-contained
python-build-standaloneif it's missing. - Creates an isolated virtualenv at
~/.rapid-mlx— nothing touches your system Python. - Symlinks the
rapid-mlxbinary into~/.local/bin. - No
sudo. Read the script first at rapidmlx.com/install.sh if you prefer.
Pin a specific version or track the latest pre-release:
$ curl -fsSL https://rapidmlx.com/install.sh | bash -s 0.10.5
Other ways to install (pip · from source)
Already manage your own Python 3.10+ environment? Install straight from PyPI:
$ pip install rapid-mlx
Add TTS / STT support with the audio extra, or build from source — the full repo, including the eval framework, is on GitHub:
$ pip install 'rapid-mlx[audio]'
Then — serve a model in one line
# auto-downloads on first run $ rapid-mlx serve qwen3.5-4b-4bit ⚡ serving on http://localhost:8000/v1
That's an OpenAI-compatible endpoint. Point Cursor, Claude Code, Aider, or any
OpenAI client at http://localhost:8000/v1 and you're running locally.