muse-glimmer

Serve Muse Glimmer locally. Development preview.

Supported platforms

• macOS on Apple silicon

Install and serve

You need Python 3.11+, Git, and Xcode with the Metal toolchain. The first installation needs an internet connection. From this repository:

pip install -e .
muse-glimmer pull SocialLocalMobile/Muse-Glimmer-30B-ExecuTorch-PTE
muse-glimmer serve

Use pull --variant <name> to choose a specific folder, or serve --model <path> for a local model.

The server runs at http://127.0.0.1:8000. Press Ctrl+C to stop. The terminal shows a live HUD.

Run Muse (coding agent) with the local model

With Muse installed, run this from your project directory:

muse-glimmer serve --launch muse

Muse uses the local model through its OpenRouter provider. No API key is needed. Exiting Muse also stops the server.

Smoke test

With the server running, open another terminal:

curl --fail-with-body http://127.0.0.1:8000/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "muse-glimmer-30B",
    "session_id": "demo",
    "messages": [
      {"role": "system", "content": "Reasoning strength: low"},
      {"role": "user", "content": "Say hello."}
    ],
    "max_tokens": 512,
    "temperature": 0.7
  }'

A successful response contains an assistant message in choices[0].message.content. Requests appear under demo in the HUD; change session_id for another session.

Configuration

default_config.toml documents all settings and defaults. To override them, pass a TOML file explicitly:

muse-glimmer serve --config ./config.toml

Run muse-glimmer serve --help for command-line options.