AI

Meta open sources Muse AI to let you build custom smart devices

Key takeaways

  • Meta has open sourced Muse AI agent code and is actively recruiting hardware makers to build Muse into consumer devices
  • Strategy avoids capital-intensive hardware manufacturing while ensuring Muse becomes the dominant consumer AI platform
  • Open source approach generates ecosystem value and user data collection opportunities that benefit Meta's advertising business

Meta is open sourcing the code behind Muse, its new AI agent platform, and asking developers and hardware makers to build their own devices around it. According to TechCrunch, the company wants Muse inside everything from smart displays and televisions to kitchen appliances and even toasters. Rather than compete with Apple and Amazon by building a whole hardware ecosystem itself, Meta is handing out the tools and betting that wide adoption will pay off for its AI business.

An "agent" here means an AI system that does tasks for you rather than just answering questions. Open sourcing means the code is published so others can inspect, change and build on it. Meta has suggested product categories including smart displays, E Ink colour screens, kitchen appliances, televisions and toasters. The vision is that within a few years Muse becomes the agent you talk to across dozens of devices at home and on your person.

What has Meta actually announced?

Meta has released the Muse agent code and is actively recruiting hardware makers to put it into consumer products. Third-party manufacturers can license Muse and build it into their own devices, instead of Meta controlling both the software and the hardware, which is how Apple works with Siri and the iPhone.

The pitch to manufacturers is simple. You get a capable AI agent without building one from scratch, and you can ship a product that talks, listens and handles tasks on day one. The pitch to Meta's investors is that every one of those devices becomes another place where people use Muse.

For background on the agent itself, see our coverage of Meta's Muse agent and what it wants access to, and of how Muse is coming to smart glasses.

Why is Meta giving the code away?

On the surface, giving away AI code looks odd. But Meta's business is not selling software licences. It is data, advertising reach and user engagement. The more devices Muse runs on, the more touchpoints Meta has with people, the more it learns about behaviour and preferences, and the more chances it has to serve personalised advertising or recommendations.

There is also a platform effect. If Muse becomes the dominant agent on consumer devices, developers will optimise for it, third parties will build integrations, and an ecosystem of complementary services will grow around it. Meta benefits from sitting at the centre without paying the full cost of hardware development and distribution.

The release also builds goodwill with developers and manufacturers at a time when Meta's public image has been damaged by privacy concerns, content moderation failures and accusations of market abuse. Presenting Muse as a collaborative, developer-friendly platform lets Meta cast itself as a good actor in AI.

How does this differ from Meta's earlier hardware efforts?

This is a calculated departure. Meta spent billions on Portal devices, Ray-Ban glasses and other hardware that largely failed to gain consumer traction, according to the report. Open sourcing Muse lets it avoid the capital and manufacturing complexity of building hardware while still capturing the value of having Muse agents everywhere.

Not every Meta hardware product has struggled, and the company still sells glasses of its own. Our look at Meta's cheapest smart glasses launch covers the price-led approach it takes there. The open Muse strategy sits alongside that, rather than replacing it.

Every time someone uses Muse on a third-party device, they are in effect using Meta's AI infrastructure. That is the trade at the heart of the plan: manufacturers get the intelligence, and Meta gets the reach and the data.

What are the technical and privacy implications?

Muse is designed to be relatively lightweight compared with larger AI models. That makes it feasible to run on consumer devices without constant cloud connectivity or high-end processors, a practical advantage over systems that need powerful GPUs or an always-on internet connection. For a rough idea of what running a large agent locally involves, our piece on Muse Glimmer hardware requirements sets out the demands.

There is a catch for privacy. The Muse agent will still be trained on Meta's infrastructure and will ultimately send data back to Meta's servers for learning and improvement. Anyone buying a Muse-powered toaster or television should expect that, and should check what the device collects before switching it on.

How do the rival approaches compare?

The competition is no longer just between individual companies. It is between competing visions of how AI fits into daily life. The table below summarises the approaches described in the report.

CompanyApproachPosition
MetaHorizontal and distributed: Muse on every deviceOpen sourced, licensed to third-party makers
AppleVertical and controlled: hardware, software and services togetherClosed, one company owns the whole stack
AmazonExisting installed base and retail relationshipsAlexa, fairly limited next to the latest AI agents
GoogleExisting installed base and retail relationshipsExpected to release or expand its own agent offerings
OpenAIDot agent platform aimed at enterprise usersClearly eyeing consumer applications

Competitors will respond. OpenAI has released Dot, its own agent platform positioned at enterprise users and businesses, though it is clearly eyeing consumer uses too, as we explain in our report on OpenAI's Dot agents moving from enterprise to consumer tasks. Google is expected to release or expand its own agent offerings. Amazon already has Alexa, though it has remained fairly limited compared with the latest generation of agents.

Samsung, LG and other appliance makers now have to decide whether to embrace Muse, build their own agents, or wait to see which platform wins.

What should developers and hardware makers know?

For hardware makers, the appeal is obvious: sophisticated AI without building it from scratch. For software developers, building for Muse means tapping into a potentially large user base. Meta will likely provide APIs, documentation and support to make integration straightforward, although none of that has been confirmed in the report.

The catch is that you are betting on Muse becoming the dominant platform. If a different agent architecture wins, your investment in Muse integration becomes a sunk cost. Given Meta's resources and distribution power, the bet looks reasonable for most manufacturers, but it is still a bet.

What happens next?

The first real test will be which manufacturers sign up, and for what kind of product. A television or smart display is a natural fit. A toaster is a more playful example, and it shows how far Meta hopes the idea can stretch.

Watch for rival responses from Google and OpenAI, and for whether appliance makers choose Muse over building their own. Watch too for how Meta handles data collection, because that is where the strategy and the public's trust are most likely to collide.

More from Future Technology HQ