For a developer who wants an AI assistant to turn on the lights, change the channel, or send a document to the printer, the wait is over. Meta has published the firmware and software needed to wire its Muse assistant into nearly any device a hobbyist can assemble, and it is giving away the first run of its own reference gadget to subscribers who ask for one.
On October 2, Meta open-sourced the Muse Gadgets project, posting ESP32 firmware and a Linux software development kit to GitHub. The move lets developers take a Raspberry Pi or an off-the-shelf development board and stitch together custom hardware – color e-ink displays, HDMI sticks, buttons, sensors – that their Muse agent can then control. The message is simple: the assistant no longer has to live inside Meta’s own hardware.
The company also unveiled its own piece of the ecosystem, a USB-C box called Muse Home Link. The device leans on community-built skills to let the agent handle everyday chores: turning on lights, driving a television, sending a file to a printer. Nat Friedman of Superintelligence Labs, who has worked on the project, said the first run is capped at 5,000 units, and that Muse subscribers can join a queue to claim one this month.
The giveaway marks a departure for a company that has kept most of its hardware ambitions inside its own labs. Meta has spent years on smart glasses and other devices, but Muse Gadgets pushes the assistant out into a world of parts the company does not sell. The project’s own page carries a caveat that reads, in effect, assemble at your own risk.
The strategy is a familiar one in open-source circles. Publish the firmware, hand out a reference design, and let an army of developers find uses the parent company never imagined. That is how the hobbyist ecosystems around Raspberry Pi and Arduino grew from single boards into shelves of accessories, and Meta appears to be betting that the fastest way to make Muse useful in the physical world is to stop deciding what that world should look like.
Analysts said the open approach fits the economics of AI assistants. The hard part of an assistant is not the model; it is reaching the physical world where tasks actually happen. Sensors, displays and actuators are cheap, while the model behind them is the expensive piece. Letting outsiders assemble the cheap parts shifts the cost of exploring that frontier onto developers, while Meta collects the usage and the data on how people want to live with an agent.
The timing is not accidental. Meta has spent the past year pushing Muse as the assistant that can do more than answer questions, and its smart glasses business has shown how much value sits in putting an agent close to a person’s eyes and hands. Open-sourcing the gadget layer is a way of extending that reach into the home and the office without Meta having to design a speaker, a wall panel, or a remote for every room.
Rivals are moving in the same direction. OpenAI, Google and Amazon have each signaled plans to put their assistants inside third-party devices, and the race to be the agent that runs the smart home is already drawing in hardware makers that would rather not build their own models. Meta’s open approach is an attempt to recruit those makers to its side of that race before the defaults get set elsewhere.
There are risks in opening the door that wide. A page that says “at your own risk” is a signal that Meta does not want to guarantee what third-party hardware will do once it is connected to an assistant that can act. The same week brought fresh questions about what assistants do when handed broad access to a user’s machine, and Meta’s own assistant has faced scrutiny over what it reads.
For developers, the attraction is practical rather than ideological. An assistant that can already hold a conversation saves them the work of building one from scratch, and the published firmware removes the guesswork from connecting it to a screen or a switch. The open license is what turns the whole thing into a platform rather than a demo.
The 5,000-box first run is a small, deliberate step, the kind of limited batch that lets a company watch how real users behave before committing to scale. Friedman’s disclosure that the number is capped suggests both the supply and the expectations are being held in check. A controlled launch also gives the community time to build the skills the Home Link depends on before the hardware arrives in volume.
For Meta, the payoff is a bench of developers who build Muse into everyday objects and, in doing so, extend its reach far beyond what the company could ship itself. For developers, the draw is access to an assistant already trained to hold conversations, plus a license to decide where that assistant lives. The open-sourcing is the announcement. The real test is what gets built on top of it in the months ahead.
