Meta is bringing its newest agentic AI model closer to the machines where people already work.
Muse Glimmer is a 30-billion-parameter open-weight model released by Meta Superintelligence Labs on Aug. 10 under the Apache 2.0 license. Designed to run locally on compatible Macs and PCs, the model can operate without relying entirely on cloud-hosted infrastructure.
Model weights are already available through Hugging Face. Developers can begin testing the release outside the company’s own evaluations.
Muse Glimmer handles multi-step agent work
According to Meta’s announcement, the agentic model can work through sequences of actions and recover when a tool produces an unexpected result. With function calling, it can interact with software during those workflows.
Multimodal support lets the AI agent process screenshots and documents alongside text. Meta cites scheduling and file organization among possible tasks. A context window of more than 131,000 tokens gives the model room to handle longer instructions and larger amounts of information.
Additionally, adjustable reasoning levels enable developers to control how much processing the model applies to a task. Support for more than 100 languages and agent frameworks such as OpenClaw adds another option to the growing field of agentic AI systems.
Local AI moves onto PCs and workstations
Hardware requirements set some boundaries around where the model can run. Meta’s official model card lists roughly 64GB of VRAM for the full-precision version. Quantized configurations target systems with around 24GB or 32GB of memory.
The social media giant is working with AMD, Arm, Dell, Intel and Nvidia on hardware optimization. Meta said integrations with tools including Ollama and LM Studio would arrive in the coming days. The company also identified vLLM and SGLang as options for larger serving environments.
Customers evaluating local AI hardware will need enough computing capacity for the model configuration they choose. Glimmer therefore brings hardware requirements directly into deployment planning instead of leaving the decision entirely at the software level.
AI agents create ongoing service work
Once an agent is deployed inside a customer environment, responsibility extends past installation. Deployments could create integration and maintenance work for MSPs and systems integrators. Human oversight also needs to be defined as part of the deployment.
Agent permissions become another consideration when the model can access files or call external tools. Meta recommends application-specific safety testing and human confirmation before irreversible actions. MSPs and MSSPs could support access controls and monitoring.
Glimmer fits into the growing AI services work handled by MSPs as customers put more AI workloads on systems they control. Providers that support those deployments can carry the work from initial integration into security and ongoing support.
Read more: Muse Code gives IT partners another coding agent to assess for deployment, integration, and governance work.