Meta's Muse Gadgets SDK turns physical AI into a device integration problem instead of a hardware program. The Apache 2.0 code targets ESP32 boards and Linux SBCs through separate device SDKs, with common surfaces for displays, audio, sensors, actuators, and buttons.
That boundary matters more than the agent. Builders can prototype on a Raspberry Pi 5, an ESP32-S3 display, or a color e-ink terminal without writing a new cloud bridge for every enclosure. The repository also includes skills for ESPHome devices, printers, TVs, and 3D printers. Hardware becomes another callable interface, while the same agent workflow survives a board swap.
The lock-in sits above the device layer because pairing still requires a Muse token and app. Embedded teams should spend one week testing whether the open SDK remains useful behind a replaceable agent adapter. If it does, vendors whose physical AI strategy requires proprietary hardware lose the prototyping layer within 12 months.