pi-mlx-models

Local MLX model launcher for Pi on Apple Silicon.

pi-mlx-models adds a local OpenAI-compatible provider to Pi, with a guided startup flow for MLX models.

What you get

• Interactive preset selector via /mlx-start
• Local runtime bootstrap via /mlx-init (venv + mlx-lm)
• Startup progress with stage/status updates and warm-up timer
• Provider models only appear when the server is actually ready
• Stop/cleanup command via /mlx-stop

Install

From npm

pi install npm:pi-mlx-models

From GitHub

pi install git:github.com/vmarinogg/pi-mlx-models

Quick start

/mlx-init
/mlx-start

When /mlx-start is used with no args, a selector opens and you can pick a preset with arrow keys + Enter.

You can also start directly with a full model id:

/mlx-start mlx-community/Qwen3-4B-Instruct-2507-4bit

Commands

• /mlx-init — initialize local MLX runtime
• /mlx-start [hf-model-id] — open selector (no args) or start with full model id
• /mlx-stop — stop server and clear active provider models

Included presets

• deepseek_r1_1_5b — reasoning, math
• gemma4_e2b — writing, summarization
• llama3_2_3b — chat, rewriting
• qwen3_4b — coding, reasoning
• smollm3_3b — fast chat, quick drafts

[!TIP] Want to try other models? Browse MLX Community collections on Hugging Face:
https://huggingface.co/mlx-community/collections

Requirements

• macOS on Apple Silicon
• Python 3.10–3.13

Configuration

Environment variables supported by the extension:

• PI_MLX_MODELS_PORT (default: 11434)
• PI_MLX_MODELS_HOST (default: 127.0.0.1)
• PI_MLX_MODELS_BASE_URL (default: http://<host>:<port>/v1)
• PI_MLX_MODELS_DEFAULT_MODEL (default: mlx-community/Qwen3-4B-Instruct-2507-4bit)

Local data paths

Runtime artifacts are stored under:

• ~/.pi/agent/pi-mlx-models/venv
• ~/.pi/agent/pi-mlx-models/models

Development

npm install
npm run build

Pi package manifest is in package.json:

• pi.extensions: ["dist/index.js"]

Screenshots

Startup progress

Command menu

Model selection