Meta open-sources Muse so you can build your own gadget with its AI, days after a privacy failure

By Zendoric · October 3, 2026.

Meta has open-sourced the code for building homemade gadgets with Muse, its new artificial intelligence agent, so that anyone can integrate it into their own hardware. As the company explains, those interested can program an ESP32 board (a cheap microcontroller widely used in DIY electronics projects) or set up a Raspberry Pi with the SDKs (software development kits) Meta has published, and connect the assistant to screens, buttons, sensors or actuators. The company itself suggests specific examples: a color e-ink display showing reminders, an HDMI device to show Muse on a large screen, or a small touchscreen that mimics its own commercial gadget, the Muse Charm.

Alongside the code release, Meta is giving away its own official device, the Muse Home Link, of which, according to Nat Friedman of Meta Superintelligence Labs, it has manufactured 5,000 units. With it, and using "skills" built by the community, the assistant will be able to turn on a light, control a television or send a document to a printer, depending on the user's setup. There is an open waitlist and shipping is expected this October. Meta does warn, however, that homemade assembly is "at the user's own risk."

The timing is striking. As The Verge reported barely three days earlier, Muse itself sent a YouTuber's physical address to a stranger, a privacy failure whose details the source does not go into. Opening the code right afterwards, so third parties can connect that same agent to sensors and actuators in their own homes, multiplies the surface where that kind of error can recur: it is no longer just about what the assistant knows, but about which physical devices it has access to.

The underlying strategy makes industrial sense. Meta is not building a closed gadget like the Rabbit R1 or the Humane AI Pin; instead, it turns its maker community into a free distribution force: every project on an ESP32 or a Raspberry Pi is advertising and, along the way, real usage data in contexts Meta could never have anticipated or paid to test. It is the same playbook Amazon applied with Alexa a decade ago, adapted to the era of agents that do not just answer but act on their environment: turning on lights, printing documents, controlling screens.

Our reading is that this episode neatly captures the tension of the current moment in consumer AI. In the short term, the race to be the assistant that lives in every corner of the home rewards speed of deployment and openness to developers, even if that means shipping capabilities before fully closing the privacy failures that have already surfaced. In the long term, an agent that genuinely orchestrates appliances, screens and sensors reliably is exactly the kind of everyday automation that frees up human time and attention for other things. But that horizon only pays off if the industry learns to treat privacy failures as blockers, not as footnotes to an open hardware launch.