Run local AI models on your Mac with LM Studio

LM Studio downloads, runs and serves local models, and on a Mac it can use Apple's MLX framework.

1. 1Go to lmstudio.ai/download and download the Mac version.
2. 2Open the downloaded file, drag LM Studio into Applications, then open it.
3. 3Open a terminal and run lms --help to check the command line tool.

You should see the LM Studio window, and a list of lms commands in the terminal.

The lmstudio.ai home page now leads with Bionic, a separate agent app. This guide uses the LM Studio app.

MLX builds are made for Apple silicon, and LM Studio lists 7 GB of memory as the minimum for this model.

1. 1Press ⌘ + 2 to open the Discover tab, then search for qwen3.5 9b.
2. 2Pick the MLX build at 4-bit and start the download.
3. 3Or skip the app and run the command below in a terminal.

lms get qwen/qwen3.5-9b --mlx

You should see Qwen3.5 9B listed in the My Models tab once the download ends.

Coding agents send long prompts, and Qwen advises at least 128K tokens of context to keep the model's thinking intact.

1. 1Run lms ls and copy the model key, such as qwen/qwen3.5-9b.
2. 2Load it with the command below. On a 16 GB Mac, use 32768 instead of 131072.
3. 3In the Chat tab, ask a short question to test it.

lms load qwen/qwen3.5-9b --context-length 131072

You should see a reply in the Chat tab, often after a short thinking section.

Qwen3.5 thinks before it answers by default. The model's Enable Thinking setting turns that off for quick replies.

Qwen suggests lower randomness for precise coding than LM Studio's defaults for this model.

1. 1Open the My Models tab and click the gear icon next to Qwen3.5 9B.
2. 2Set Temperature to 0.6 and Presence Penalty to 0.
3. 3Leave Top P at 0.95 and Top K at 20, then save.

You should see the new values in the model's default settings, used every time it loads.

These are Qwen's numbers for coding with thinking on, from its model card. An app that sends its own temperature overrides them.

The server gives the model an OpenAI-compatible address that coding tools can call.

1. 1Open the Developer tab and turn on Start server. Or run lms server start in a terminal.
2. 2Run the command below to check that the server answers.

curl http://localhost:1234/v1/models

You should see a JSON list that includes qwen/qwen3.5-9b.

The server uses port 1234 and needs no API key by default. It stays on this Mac unless you choose to serve on your local network.

With a local model behind them, both agents help you code for free, and your code never leaves your Mac.

1. 1OpenCode: install it from opencode.ai/docs, then run opencode in your project folder.
2. 2Type /models and pick Qwen3.5 9B under LM Studio. OpenCode finds LM Studio at http://127.0.0.1:1234/v1 on its own.
3. 3Pi: add the config below to ~/.pi/agent/models.json, run pi, then type /model and pick the model.

{
  "providers": {
    "lmstudio": {
      "baseUrl": "http://localhost:1234/v1",
      "api": "openai-completions",
      "apiKey": "lm-studio",
      "models": [
        { "id": "qwen/qwen3.5-9b", "reasoning": true }
      ]
    }
  }
}

You should see the agent answering in your project with Qwen3.5 9B as its model.

Pi's models guide lists every field. Never put a real cloud API key in this file if you share it.