# Muse Glimmer

An offline-first interactive lab and long-form guide to Muse Glimmer, Meta's open-weight 30-billion-parameter agentic multimodal model, letting a reader explore agent loops, benchmarks, and memory sizing before downloading the model weights.

This repository is a companion lab and guide to Muse Glimmer, an open-weight, 30-billion-parameter agentic multimodal model released by Meta Superintelligence Labs on 10 August 2026 under Apache 2.0, optimized for always-on local agent workflows and runnable on consumer hardware with roughly 20 GB of memory after 4-bit quantization. It does not itself contain the model weights, which are published separately.

A long-form introduction (`BLOG.md`) and an offline-first interactive lab let a reader feel out agent loops, benchmarks, and memory envelopes before downloading anything; the lab and its benchmarks work with no GPU and no API key. Optional live chat can connect to a locally served GGUF model through llama.cpp or to a Hugging Face Inference Endpoint, following a separate live-setup guide.

Running the lab locally needs Python 3 and the repository's pinned requirements, or a provided Docker Compose file; both start a local web server on port 7870.

Website: https://github.com/cobusgreyling/Muse-Glimmer