- Published on
Everyone Can Have Their Own AI Agent Device: Meta Open-Sources Muse Gadgets for Building Custom AI Hardware with ESP3...
- Authors
- Name
- DeepHorus
On October 2 local time, Meta announced the release of the open-source Muse Gadgets project, enabling developers to build custom hardware for its personal AI agent, Muse. The project has been published on GitHub under the Apache 2.0 license and includes ESP32 firmware and a Linux device SDK. Developers can connect a low-cost ESP32 development board or a Linux device such as a Raspberry Pi to Muse and hook up a display, buttons, sensors, actuators, and other network-capable devices on the workbench.
In terms of hardware support, Meta’s ESP32 SDK documentation currently covers 11 development boards, including Espressif’s ESP32-C5 DevKitC-1 (the official recommendation), M5Stack StickS3, the Waveshare touch AMOLED display, and Seeed e-paper displays, among others. Six of these boards can run the complete screen UI, including an animated avatar, button-based calling, and settings. Boards with enough memory also support home-network tunneling, allowing Muse to access devices the user already owns, as well as any custom-built devices that expose a local HTTP API.
On the Linux SDK side, Meta now supports the Raspberry Pi 3B+, 4, 5, and Zero 2 W, along with other Linux devices featuring Bluetooth Low Energy support. The minimum supported system requirements are Raspberry Pi OS Bullseye, Debian 11, and Ubuntu 22.04. According to the documentation, Muse executes commands as the account selected during installation and has all of that account’s permissions: “If that account can use sudo, so can Muse.” The installer asks for confirmation before handing over those privileges, and developers can also use install.sh --run-as to run Muse under a specified unprivileged account. On Linux devices, Muse can execute shell commands, read and write files, and check device health. Official examples include installing Home Assistant on a Raspberry Pi and checking every morning at 7:00 a.m. whether the backups have run.
Meta has also used the SDK to create a finished device called Muse Home Link. Nat Friedman, product lead at Meta Superintelligence Labs, described it as a compact, USB-C-powered device that allows Muse to connect to a home network and communicate with televisions, speakers, and other smart devices that expose HTTPS interfaces. Friedman said Meta produced approximately 5,000 Home Link units in its initial production run and is offering them free of charge to Muse subscribers in the United States, with one device per subscriber on a first-come, first-served basis.
The launch of Muse Gadgets is part of Meta’s recent rapid expansion of the Muse ecosystem. Earlier this week, Meta also launched Muse for Small Business. Its free tier includes usage limits, while its $20 or $100 monthly paid plans can connect to third-party tools such as Shopify, Dropbox, Slack, Canva, Figma, and QuickBooks. Meta has also established a new business unit called Meta Enterprise Platform, led by former MongoDB CEO CJ Desai. It brings together products including the Muse agent, Meta Business Agent, Muse API, and Muse Code to promote AI tools to business customers.
From an industry perspective, Muse Gadgets marks Meta’s effort to expand Muse from a simple chatbot into an agent runtime that operates across different types of hardware. If third-party hardware only needs to expose standard interfaces for Muse to understand and invoke it, personal AI agents may eventually take on some of the roles traditionally performed by operating systems, applications, and smart-home hubs.
Nevertheless, Meta also provides a notably candid risk warning in the SDK documentation. Community pairing “has no manufacturer verification and cannot prevent active man-in-the-middle attacks,” so pairing should be completed on a trusted network. SDK tokens are built into the firmware and should be treated as identifiers rather than passwords; if one is exposed, it must be revoked and the firmware rebuilt. Wi-Fi credentials and device tokens are stored in flash memory, so NVS encryption is strongly recommended. Build artifacts are signed with development keys, and secure boot is disabled by default. The documentation concludes: “The side effects of tinkering may include bricking the device, voiding the warranty, voltage drops, or going broke. You assume the risks.”
