Meta Open-Sources Muse Gadgets for ESP32 and Pi, With 5,000 Free Home Link Units
Meta is letting users build their own Muse gadgets around its Muse AI agent, using code it has open sourced. The Verge
Meta describes Muse gadgets as open-source devices users assemble themselves. The model is bring-your-own hardware plus Meta-supplied software, not a retail product line. You buy the board, you set it up.
There are two supported build paths. You can program an off-the-shelf ESP32 board, a small and cheap microcontroller widely used in DIY electronics, using Meta's SDKs, the software toolkits for building on that hardware. Or you can set up a Raspberry Pi, a low-cost, credit-card-sized computer popular with students and hobbyists, using the same SDKs. In both cases the builder sources the board and handles provisioning, the initial setup that connects it to software and networks.
Meta points to two example builds to show the range. One puts Muse on a color E Ink display, the low-power screen technology used in e-readers, to show reminders. The other puts Muse on an HDMI stick, a small plug-in computer that drives a TV, to show Muse on a big screen. One is small, persistent and glanceable. The other is shared and large-format.
Alongside the DIY release, Meta built its own reference device. It is called Muse Home Link, and the company is giving it away rather than selling it.
Meta made 5,000 Muse Home Link units, according to Meta Superintelligence Labs' Nat Friedman. That number sets the size of the initial distribution.
Muse Home Link uses community-built skills, small add-on programs that extend what the Muse agent can do in the home. Depending on setup, those skills let it handle tasks such as turning on a light, controlling a TV, and sending a document to a printer. What it can do depends on which skills are installed and which devices are paired.
Meta opened a waitlist for Muse Home Link ahead of shipping later in October. The details were published on Oct. 2, 2026, which puts shipment later that same month.
The broader context here is how Meta is dividing effort and risk. A 5,000-unit run does not supply a market. It seeds a developer base. By pairing a limited reference device with SDK support for commodity ESP32 and Pi hardware, Meta avoids tooling, inventory and support for a consumer launch while still getting physical endpoints into homes where integration problems are real and specific.
In my view, the skills dependency deserves close attention from anyone planning to build. Turning on a light, switching a TV input or routing a document to a printer looks like a simple voice command in a demo. In a lived-in home network it involves discovery, credentials, permissions, timeouts and inconsistent device implementations. Community-built skills can cover that long tail quickly. They also push testing, maintenance and trust decisions out to individual builders and users, with direct consequences for reliability and security.
Looking at what this enables if builders take it up, the two examples matter more than they first appear. An E Ink reminder panel suggests an always-on, low-attention surface with minimal interaction cost. An HDMI stick suggests a shared display where agent output is legible to a room, not just one user. Both move the agent off the phone and laptop without requiring new silicon. For practitioners, the useful question is architectural: which agent tasks actually benefit from dedicated, single-purpose hardware, and whether open client code on cheap boards is enough to answer it at scale.
Worth flagging for longer-term expectations, this pattern is familiar from earlier waves of extensible hardware. Small reference runs plus open SDKs tend to produce a burst of inventive prototypes, followed by a harder phase around upkeep, compatibility and user support. That second phase is where the value is decided. If the skills ecosystem gets versioning, clear permission boundaries and straightforward debugging, these gadgets could become durable infrastructure for experimentation. Raising two kids taught me that tolerance for fiddly setup is low, even among enthusiastic early adopters, but tolerance for a device that quietly does one job well is very high.
