Meta releases Muse Glimmer for local AI agents on consumer hardware
Meta has released Muse Glimmer, a 30-billion-parameter open-weight AI model designed to run agents locally on consumer hardware. The company says Glimmer can operate on a Mac or PC with a single consumer GPU, process text and images, and support more than 100 languages.
Glimmer is an open version of Meta’s more powerful closed model, Muse Spark, which debuted in April. Its weights are available under the permissive Apache 2.0 licence, allowing developers to download and modify the model for their own uses.
A model for extended local workflows
Meta designed Glimmer for agents that perform multistep work rather than respond only to a single prompt. The model can call tools, write and debug code, work with files and screenshots, and execute tasks across an extended workflow.
The intended use cases include managing schedules, drafting messages and organising files. Those activities can require access to substantial amounts of personal information, and Meta presents on-device processing as a more privacy-sensitive alternative to sending that information to the cloud.
Glimmer is also intended to be always on and usable anywhere, anytime, including without an internet connection. That combination of local execution, tool use and persistent availability frames the model as infrastructure for a personal agent rather than a conventional cloud chatbot.
Open weights, with a clear limit
The release gives a practical indication of how Meta is separating models it distributes from models it retains under company control. In Meta’s agentic coding portfolio, Meta’s Muse Spark 1.1 coding model remains a closed-weight model, while Glimmer can be downloaded, fine-tuned and operated on a user’s own hardware.
Mark Zuckerberg has argued that advanced AI should empower individuals instead of being concentrated among a small number of companies. In a letter accompanying the release, he said widely distributing superintelligence could create a new era of personal empowerment, and described agents that could work continuously on relationships, health, careers, finances, home management and hobbies.
At the same time, Meta has said it must consider safety when deciding which increasingly capable models to release openly. Glimmer therefore does not represent the full capability of Muse Spark; it establishes an accessible, adaptable tier while Meta keeps its stronger model closed-weight.
What businesses should assess
For businesses, Glimmer creates an option to evaluate locally operated agents where data handling, offline use and model modification matter. Teams should assess the permissions an agent receives, the consumer hardware needed for deployment and the difference between an open model they can adapt and a higher-capability service that remains controlled by its provider.