Meta open sources Muse AI gadget SDKs for ESP32 and Raspberry Pi

What Meta released

Meta has published the code needed to run its Muse AI agent on hardware that makers build themselves. According to The Verge, the company frames these as do-it-yourself devices: program an off-the-shelf ESP32 board or set up a Raspberry Pi with Meta's SDKs, then wire Muse into displays, buttons, sensors and actuators from your own parts bin.

Two SDKs ship in the release, per the Muse Gadgets page Meta put online alongside the code. The ESP32 Device SDK links a board to Muse and supports adding a screen, audio input and output, or other sensors. The Linux Device SDK turns a spare Raspberry Pi or any Linux box into a Muse gadget, with room for custom commands — the site suggests letting the agent take on sysadmin chores or drive a Home Assistant installation.

Everything sits on GitHub under an Apache 2.0 license, provided as-is and without warranty. Meta's warnings are candid, noting that tinkering may result in bricked boards, voided warranties or power problems, and telling builders to proceed at their own risk.

Reference builds and boards

The page lists featured example devices, most made and sold by third parties that Meta says it neither endorses nor warrants. They include a Waveshare ESP32-S3 board with a round 1.75-inch AMOLED touchscreen, push-to-talk, a speaker, mic and battery; a Seeed reTerminal E1002 that gives Muse a color e-ink display for morning briefings, reminders and shopping lists; an M5Stack StickS3 that shows the agent's status on a pocket-sized screen; the AiPi Lite, a small desk device that answers out loud and draws pictures on its color display; and the Home Assistant Voice Preview Edition, which pairs push-to-talk with an LED status ring. A Raspberry Pi 5 build targets Home Assistant and other Linux applications, while an HDMI stick that puts Muse on a TV is listed as coming soon.

The Verge points out that some builds mirror Meta's own products — a small touchscreen device resembles a homemade Muse Charm — and that an e-ink setup can surface reminders. A write-up on dev.to argues the release is more than an API wrapper, since builders get the agent runtime and with it control over form factor and privacy boundaries; that post also claims gadgets can run with minimal cloud dependence, though Meta's own materials do not spell out how much of the agent executes locally versus remotely.

The Muse Home Link giveaway

Meta also built a reference gadget of its own, the Muse Home Link, and is giving it away. Per The Verge, it leans on community-built skills so Muse can switch on a light, control a TV, send a document to a printer and more, depending on the user's setup. Nat Friedman of Meta Superintelligence Labs says 5,000 units were manufactured. A waitlist is open now, with shipping expected sometime this month. A community Discord for builders is also live, and it showed only a few dozen members online when the page was up — a sign of how early this ecosystem is.

Why it matters

Most AI hardware today arrives sealed: you buy the device, accept the firmware as shipped, and trust the vendor. Meta is inverting that model by treating an AI companion as something you assemble from a few dollars' worth of dev boards and open source code. If the community takes to it, Muse could spread across far more form factors than Meta would ever build itself — desk toys, e-ink dashboards, TV sticks — while the Home Link seeds a skills ecosystem on top.

The catch is that an open hardware layer does not make the whole stack open. The SDKs are described as connecting boards to Muse, which suggests the agent itself still runs wherever Meta runs it, and the company has not detailed that architecture. Builders get control over their circuits, but the brain remains Meta's — a distinction that will matter to anyone drawn to this project by privacy arguments.