Meta just gave developers a real look at what Mark Zuckerberg means by “personal superintelligence.” On Monday the company released Muse Glimmer, an open-weight model built to run AI agents locally on consumer hardware, according to TechCrunch AI. It’s the clearest signal yet of where Meta wants personal AI to live: on your own machine, not in someone else’s cloud.
Glimmer is a 30-billion parameter model, essentially an open version of Muse Spark, Meta’s most powerful closed model, which the company debuted in April. The weights ship under the Apache 2.0 license, so developers can download them, fine-tune them, and modify them freely. That’s a permissive license with almost no strings attached.
What Glimmer actually does
This isn’t a chatbot. Glimmer is built to power agents that carry out multi-step tasks, and TechCrunch AI reports it can do a lot on a single consumer GPU in a Mac or PC:
- Call tools and work across an extended workflow
- Write and debug code
- Handle files and screenshots
- Support text and images, trained across more than 100 languages
Meta pictures it managing schedules, drafting messages, and organizing files. All of that needs deep access to your personal data. The pitch is that by processing everything on your device instead of shipping it to the cloud, Meta is building toward a more privacy-sensitive personal agent. Glimmer is also designed to be “always-on” and to work “anywhere, anytime, with or without an internet connection.”
Why this matters
The status quo for capable AI agents has been cloud-first. You send your data to a company’s servers, their model does the work, and you trust them with it. Running a genuinely useful agent on local hardware changes that equation. It’s a privacy story, but it’s also a control story. What stands out here is that Meta is betting individuals will want to own the intelligence that touches their most personal information.
Zuckerberg made the philosophy explicit in a letter Monday. He argued that distributing superintelligence widely “has the potential to begin a new era of personal empowerment where individuals can use this powerful new capability to reach their full potential, pursue their interests, and improve their lives and the world more than ever before.” He described a personal agent that “will work 24/7 on your behalf” across relationships, health, career, finances, and more, and promised that “everyone will have free or affordable access to these tools.”
The line Meta is drawing
Here’s the tension worth watching. Access isn’t the same as ownership. Meta is increasingly splitting its lineup between models it releases openly and models it keeps locked down. Muse Spark, the more powerful one, stays closed-weight. Glimmer, the smaller sibling, is the one you can download and run yourself.
So Glimmer marks an early boundary line. On one side sits the AI Meta wants you to own outright. On the other sits the more powerful intelligence Meta intends to keep under its own roof. Zuckerberg has said before that Meta would be careful about which of its increasingly capable models it opens up, citing safety concerns. This release shows that caution in practice: give away the 30B agent, hold back the flagship.
What to expect next
For developers and builders, Glimmer is worth grabbing now. An Apache 2.0 agent model that runs on a single GPU and handles tool use, code, and files is a strong foundation for local-first products, especially anything where sending user data to the cloud is a dealbreaker. Expect a wave of fine-tuned variants and privacy-focused personal assistants built on top of it.
The bigger question is where Meta keeps drawing that open-versus-closed line as its models get stronger. Glimmer is generous. Spark is not. As the gap between the two grows, so does the gap between the AI you can own and the AI Meta rents you. For now, the download is real and the capability is meaningful. More details are available in the original TechCrunch AI report.