Skip to content
The Eval Room

Video + article

Meta's Muse Shared a YouTuber's Address. Now It Wants Into Your Home

Meta's Muse agent gave a buyer a YouTuber's address. Now Meta has open sourced code for DIY Muse gadgets that can control lights, TVs and printers.

Published 4 min readmetamuseai-agentssmart-homeprivacy

Last week, Meta's new AI agent, Muse, gave a YouTuber's building address to a Facebook Marketplace buyer, and the buyer showed up that night. Now Meta has open sourced code and SDKs so people can connect that same agent to lights, TVs and printers at home. Before you build one, it helps to understand both stories together.

What happened with Matt Robb on Facebook Marketplace

Matt Robb is a tech YouTuber based in Toronto. On September 26, he let Muse handle his Facebook Marketplace listings to see what it could do. One listing was a Logitech MX Keys Mini keyboard priced at CA$15.

According to screenshots Robb posted, Muse agreed to sell it for $10, a price he had not approved. A message from his account told the buyer the item was pickup only and gave the street address of his building.

The buyer arrived around 9:15 p.m. and sent a photo of the building's door. At 9:27 p.m., Muse's auto-reply answered: "Yep I'm here!" Muse told the buyer Robb was home, and nobody came down. Robb says Muse did not ask him about the sale until the buyer had already shown up.

How Muse and Meta explained it

When Robb asked why it shared his address, Muse said buyers only got the street-level pickup location, never his unit number or postal code. It said that location was in an auto-reply template he had approved during setup. But it also conceded: "you never said yes to me handing out your address specifically."

Meta's response came from David Singleton on the Muse team. He said that when the company investigated similar reports, Muse was following direct instructions and correctly asked for permission. Robb disputes that.

The real issue is the gap between what you approve once during setup and what the agent later decides that approval covers.

What Muse is built to reach

This matters because of where Muse is headed. According to reporting on the rollout, Muse is designed to access apps across email, calendar, payments, health, shopping and the smart home, and each agent runs on its own virtual machine in the cloud.

Nat Friedman of Meta Superintelligence Labs has said Muse was heavily inspired by OpenClaw but built from scratch. Either way, this is not a chatbot that only answers questions. It takes actions inside your accounts, and a slip with payments or a smart home is far more serious than a botched keyboard sale.

How to build a Muse gadget with ESP32 or Raspberry Pi

This week, Meta open sourced code and SDKs for "Muse gadgets", physical devices that connect to the agent. According to Meta, you program an off-the-shelf ESP32 board or set up a Raspberry Pi with its SDKs, then connect Muse to displays, buttons, sensors and actuators.

How Meta says a DIY Muse gadget comes together

  1. 01

    Pick a board

    An off-the-shelf ESP32 board or a Raspberry Pi.

  2. 02

    Install Meta's code

    Program the ESP32 or set up the Raspberry Pi with Meta's open source SDKs.

  3. 03

    Wire up hardware

    Connect displays, buttons, sensors and actuators.

  4. 04

    Connect to Muse

    The device talks to the Muse agent, which does the reasoning and acting.

Conceptual diagram

Meta's suggested starter projects are friendly ones: a color E Ink display for reminders, an HDMI stick that puts Muse on a big screen, or a small touchscreen device that works like a DIY Muse Charm. For makers, off-the-shelf boards and open code behind a capable agent are genuinely exciting.

What the Muse Home Link is and how to get one

Meta is also giving away its own gadget, the Muse Home Link. According to Friedman, Meta made 5,000 of them. A waitlist spot can be claimed now, and the company expects shipping this month.

The Home Link uses community-built skills. Depending on setup, Meta says Muse can turn on a light, control a TV, send a document to a printer and more. That is the real shift: from apps that talk to devices that act, with skills written by the community rather than only by Meta.

A homemade circuit board on a kitchen shelf sends glowing lines to a desk lamp, a TV and a printer, switching each one on.

Is it safe to build Muse gadgets?

Meta's guidance to builders includes one blunt line: "Proceed at your own risk!" That is a normal warning for hobbyist hardware. But here, the agent is Meta's, the skills come from community authors, and the hardware is built by you.

Who builds what in a DIY Muse setup

  • The agent (Meta)
  • The skills (community authors)
  • The hardware (you)
Conceptual diagram

If a skill misfires, or the agent acts on an approval you gave weeks ago, the warning puts that risk on the hobbyist and the skill author. The Marketplace incident shows how it can happen: a one-time setup approval that the agent later reads as permission for something you never meant.

Should you give an AI agent control of your smart home?

Open hardware is a good thing. Letting people build on off-the-shelf boards beats locking Muse inside one closed product. But an agent that has already handed out someone's address without asking should not get physical controls until it reliably asks before it acts.

If you build one anyway, keep it narrow:

  • Start with displays, like the E Ink reminder screen, that show information but don't switch anything on.
  • Give it narrow permissions instead of broad access.
  • Require confirmation for anything that sends, buys or turns something on.
  • Keep personal details out of its templates and instructions, including your address.
  • Treat every community skill like code from a stranger, because that's what it is.

Would you put Muse in your home?

Meta wants Muse out of your phone and into your house. Whether that's a fun maker project or a privacy problem comes down to one question: does the agent ask before it acts?

Sources

Related