Meta Releases Open AI Model for Local Agent Execution

Meta has released Muse Glimmer, a 30 billion parameter open-weight model designed to run autonomous AI agents directly on consumer hardware. This allows agentic workloads, typically reliant on cloud infrastructure, to operate on high-end Macs and PCs.

The model is released under the permissive Apache 2.0 open source license, marking Meta's first fully open release since its proprietary Muse Spark. The Apache 2.0 license is more open than the previous custom license for the Llama family, permitting unrestricted commercial use and modification.

Muse Glimmer is specifically trained for agentic workflows, featuring capabilities for planning, tool calling, result interpretation, and error recovery. Meta has optimized the model for local inference, capable of running on systems with 24GB of VRAM, enabling local processing of sensitive data without continuous transmission to remote servers. The model also supports interleaved text and image processing.

Meta is collaborating with hardware partners including AMD, Arm, Dell, Intel, and Nvidia to optimize performance across devices. The model weights are available on Hugging Face, with support rolling out for various local runtimes. Meta is also providing developer documentation for building custom agent scaffolds.