Meta plans to sell a cute, carry-anywhere AI companion called Muse Charm. A homemade version built from cheap parts suggests that most of the value sits in the AI model behind the device, not in the device itself.
Key takeaways
- Meta plans to sell Muse Charm this holiday season as a small, carry-anywhere AI companion with a friendly animated face, in the style of a Tamagotchi.
- The Verge has published a preview describing a homemade alternative, which appears to be built on a Waveshare ESP32 board and coded with help from Anthropic’s Claude.
- Small microcontroller boards with screens and wireless connections are cheap and widely available, so the physical side of an AI companion can be copied without much difficulty.
- A commercial product may still be worth more than a hobby build if it adds safety, reliability, battery life or features that depend on its maker’s wider services.
The hardware is the easy part of an AI companion
The Verge reports that Meta has made its AI assistant look appealing by giving it a soft, expressive cartoon face. The company now wants to put that character into a small physical gadget, the Muse Charm, which it intends to sell over the holiday season. The publication compares the idea to the Tamagotchi, the egg-shaped virtual pet that people carried on keyrings and looked after through a tiny screen. The writer says they will not be buying one, because they have already made something similar at home.
The argument of this article follows from that comparison. A pocket AI companion is made of two things: a small device with a screen, a microphone or buttons and a wireless connection, and a large AI model that does the talking and thinking. The device is the part people see and touch, but the model is what makes it useful or entertaining. If someone with modest technical skills can build the device for little money in a short time, then much of what a commercial product offers, and much of what it is worth, depends on the model and the services behind it rather than the plastic shell.
That has consequences for buyers and for the companies selling such gadgets. Buyers should ask what a dedicated device does that a phone app or a home-made gadget cannot. Companies have to show that their hardware adds something beyond a cute look.
Many details of Muse Charm are not known from the material available here, including its price, its exact specifications, its battery life, whether it needs a phone nearby and which features it will have. This article does not guess at them. Its claim is narrower: the basic form of the product, a small screen with a friendly face linked to an AI assistant, is not hard to reproduce.
Cheap microcontroller boards already provide the core parts
The address of The Verge’s preview mentions Waveshare and the ESP32, which points to the hardware its writer used. The ESP32 is a family of low-cost microcontroller chips made by Espressif Systems. Hobbyists and manufacturers use them widely because they combine a processor with built-in Wi-Fi and Bluetooth on one small, inexpensive chip. Waveshare is an electronics company that sells development boards and add-ons, including compact boards that pair ESP32-class chips with small colour displays and, on some models, touch input, microphones or speakers.
With a board like that, the main physical parts of a Tamagotchi-style AI companion come in one package. The screen can show an animated face. The wireless chip can connect to the internet. Buttons or a touch panel let the user interact with it, and an audio add-on can handle voice. A case, a battery and a keyring loop complete the object. None of this needs custom chip design or factory tooling. These are off-the-shelf parts sold to the maker community.
This is why “we have Muse Charm at home” is more than a joke. Small screens and wireless chips are commodity components that many suppliers sell. A large company can make a slimmer, more durable or more attractive version, but it cannot easily stop others from building the same basic thing. Exactly how close the homemade version comes to Meta’s product in size, finish or ability is not established by the material available here.
AI coding assistants lower the software barrier
Hardware is only half the job. Someone also has to write the software that draws the animated face, records the user’s voice or reads button presses, sends requests to an AI service, receives the replies and shows or speaks them. Writing embedded software for microcontrollers has usually needed specialist knowledge of low-level programming, device drivers and memory limits.
The preview’s address also mentions Claude, Anthropic’s AI model, and “vibe code”. Vibe coding is the informal name for building software mostly by describing what you want to an AI model in plain language, accepting the code it produces, testing it and asking for changes, rather than writing every line yourself. That suggests the writer relied on an AI assistant to produce much of the software. The material available here does not say how much code was generated, how many attempts it took or how well the result works.
The general point does not depend on those details. When AI tools can produce working code for a small display and a network connection, the specialist skill needed to build a gadget like this falls sharply. A person who could not have written microcontroller firmware a few years ago can now get a working prototype with help from a chatbot. The barrier that used to separate a polished consumer product from a weekend project is lower than it was, though code made this way can still contain errors that the builder may not notice.
There is also a neat circularity here. An AI model helps build a device whose main purpose is to talk to an AI model. As a result, the hardware becomes less distinctive and more of the value moves to the AI services themselves.
The personality lives in a cloud model, not the casing
A microcontroller such as the ESP32 has very little memory and processing power compared with the computers that run large language models. It cannot run a capable conversational AI on its own. In practice, a device like this works as a front end. It captures what the user says or types, sends it over the internet to an AI model on remote servers, and displays or plays back the answer.
That is very likely true of a homemade build, and it is a reasonable assumption for any small, cheap companion gadget, though the material available here does not describe how Muse Charm handles processing. If the intelligence sits in the cloud, the device’s job is mainly to provide a screen, a face and a handy way to reach the assistant. The conversation, the knowledge, the tone and the memory all come from the model and the service around it.
This explains why the cute face matters so much to Meta’s approach, as The Verge describes it. A friendly character makes an AI assistant feel approachable and gives a physical product an identity. But the face is a design asset, and design assets can be imitated. A hobbyist can draw their own character, connect it to whichever AI service they choose and end up with a companion that looks and behaves differently from Meta’s, yet is built on the same basic structure.
The case against: commercial products offer what hobby builds lack
The strongest objection is that this argument confuses a prototype with a product. Getting something to work on a desk is very different from making something that millions of people can use every day without trouble. A commercial device has to survive being dropped, last a reasonable time on one charge, connect reliably, receive security updates and meet safety and radio regulations. Hobby builds often have exposed wiring, short battery life, fragile cases and software that breaks when an online service changes. Most people do not want to solder, flash firmware or debug code in order to own a toy.
Integration is a second objection. A product made by a large technology company may connect to that company’s accounts, messaging apps, photos and other services in ways that an independent build cannot easily match. If Muse Charm turns out to link closely with Meta’s wider products, that integration could be the real offer, and a homemade copy would not have it. Whether this is the case is not known from the available material.
A third objection concerns cost and convenience. A homemade build may look cheap in components, but it takes time, tools and some ongoing costs. Access to a capable AI model through a developer interface is often charged by usage, and the builder has to manage keys and accounts. A finished product sold with its service included hides that complexity. For many buyers, paying for something that simply works is the sensible choice.
Taken together, these points mean the hardware may be easy to copy while the product is not. A device sold to the public involves manufacturing, support, safety and integration, and none of these is cheap.
Evidence that would change the conclusion
Several developments would weaken the claim that the gadget matters less than the model. If Muse Charm runs a meaningful part of its AI processing on the device itself, offering offline use, faster responses or better privacy, its hardware would be doing real work that a basic microcontroller build could not copy. Custom chips or sensors that cheap boards lack would have the same effect.
Exclusive features tied to the physical product would also count. If the device can do things that are only available through Meta’s own hardware, such as particular kinds of awareness of its surroundings, close links to other devices or capabilities that the company does not offer to outside developers, then the gadget would be more than a container for a cloud service.
On the other side, the conclusion would get stronger if detailed reviews show that Muse Charm mostly passes requests to a remote model with little processing of its own, or if hobbyist builds turn out to match it closely in everyday use. A full public account of how the homemade version was built, how long it took and how reliable it is would also help show how low the barrier has fallen. Until the product’s specifications and pricing are published and independent testing is available, the comparison remains a reasoned argument rather than a settled finding.
Sources and further reading
- The Verge, gadget coverage previewing Meta’s Muse Charm and describing a homemade alternative
- Espressif Systems, technical documentation for the ESP32 family of microcontrollers
- Waveshare, product listings and documentation for development boards with integrated displays
- Anthropic, public documentation describing Claude and its use for writing code
Surfaced from the rss:verge signal “DIY AI companion gadget”. AI-assisted draft, editorially reviewed.
