This week, Meta's Muse agent topped the US App Store productivity chart, with 100K+ ratings and 4.9 stars. In the same period, Meta open-sourced the Muse Gadget SDK on GitHub, letting developers assemble a desktop AI physical assistant for the cost of a few dozen yuan of ESP32 dev board hardware. Beyond the chart-topping itself, we care more about the other side — AI is stepping out of phone screens, and the path of last-generation hardware (Rabbit r1, AI Pin) trying to replace phones has already failed.

What this is

Muse is Meta's personal AI agent, and this week it surged to #1 on the productivity chart. What truly drew attention was the simultaneous open-source release of the Muse Gadget SDK, including complete edge-cloud code targeting ESP32 (IoT chips starting at a few dollars) and Linux (embedded devices like Raspberry Pi).The architecture adopts a "light edge, heavy cloud" approach: ESP32 only handles audio capture, buttons, and GPIO control (manipulating physical devices through general-purpose input/output pins). LLM inference, memory retrieval, and tool calls all happen in the cloud. Devices maintain real-time connections with the cloud via WebSocket (a long-connection communication protocol), with audio encoded in Opus format (an efficient audio codec) for real-time transmission. Tool calls are dispatched through an MCP-like protocol (an open standard letting AI invoke external tools). Developers only need a dev board, a microphone, and a speaker to build a desktop AI assistant — hardware cost dropped to under a few dozen yuan.

Industry view

Supporters view this as a direct response to the failures of Rabbit r1 (starting at $200) and Humane AI Pin — last-generation AI hardware tried to "replace the phone," but users who already carry an iPhone won't carry a second device. Muse positions itself as "desktop-resident, environment-aware," complementing rather than competing for the entry point. But opposition is equally clear: first, open-source SDK does not equal consumer product — industrial design, supply chain, and after-sales support from dev board to usable hardware remain unsolved challenges for Meta. Second, ESP32 can't run any local model; every wake-up requires network connectivity, and offline means a bricked device. Privacy risks scale with adoption. Whether "desktop-resident + always-on microphone" passes users' privacy bar is the real test — many already have reservations about phone and smart speaker microphones; one more always-listening environmental device is a different psychological burden.

Impact on regular people

For the consumer market: the next 12-18 months will see a wave of Muse-form-factor hardware in the 100-500 yuan range ($15-$75), but most will die on industrial design or supply chain. For personal careers: white-collar workers shouldn't rush to buy, but should recognize that AI entry points are spreading from apps to resident physical environments — voice interaction density on future office desks will noticeably increase. For enterprise IT: the open-source SDK gives traditional hardware vendors a low-barrier entry point, but they need to decide first whether they're building "consumer products" or "developer kits" — the business models are completely different.