Muse Glimmer Starter Script

Set up and run Muse Glimmer 30B locally on an Apple Silicon Mac with
llama.cpp and
OpenCode.

The starter script downloads the model, starts a local server, and connects it
to OpenCode. Everything runs on your Mac and the server listens only on
127.0.0.1.

Requirements

• An Apple Silicon Mac. We recommend 64 GB of unified memory; the first model
  download needs about 30 GiB of free disk space.
• git, curl, Python 3.10+, a recent llama-server, and OpenCode 1.x.
• No pip install is required.

Add the installed commands to your shell:

echo 'export PATH="$HOME/.local/bin:$PATH"' >> "$HOME/.zshrc"
export PATH="$HOME/.local/bin:$PATH"

Quick start

Clone this repository, open it in Terminal, and run:

make install    # install the muse-* commands
make download   # download and verify the model files
muse-start      # start the local model server
muse-code       # open OpenCode with Muse Glimmer

The first download is about 21 GB and may take a while. When ready,
muse-start reports that the server is healthy at http://127.0.0.1:8080, and
muse-code opens OpenCode using the local model. Later runs reuse the verified
model files.

More details

See the reference guide for custom llama.cpp locations,
all available commands, runtime selection, configuration, logs, testing,
updates, troubleshooting, and uninstall instructions.

License

The starter scripts are licensed under the Apache License 2.0. Model
weights are downloaded separately and remain subject to the terms published by
the upstream
Muse Glimmer 30B GGUF repository.