Nearly three decades after the Tamagotchi turned a virtual pet into a pocket companion, Meta has revived a similar aesthetic for a very different purpose. Presented at Meta Connect 2026, the Muse Charm places the personal agent Muse inside a small device, allowing it to interact with the user via voice without opening an app on their phone. The nostalgic look catches the eye, but it is what lies behind it that makes the release more interesting. Muse can access authorized services, execute tasks, and use information from conversations to customize its responses and actions. In just a few weeks, its app accumulated millions of downloads. The proposal raises a question: what information and permissions does an AI need to receive to execute tasks on our behalf?

Muse Charm: what is Meta's “AI Tamagotchi” and how does it work

Presented during Meta Connect 2026, the Muse Charm is a small device created to take the personal agent Muse off phone and computer screens. According to Meta, the device allows users to chat with Muse using a real-time voice model.

Despite comparisons to a Tamagotchi, Muse and Muse Charm are not the same thing. Muse is the artificial intelligence agent capable of conversing and performing tasks. The Charm works as a new physical interface to access this agent. In the demonstration presented by the company, the device has a small screen that displays Muse's avatar and a fingerprint sensor that allows voice interaction to begin without needing to unlock the phone or open an app.

The goal is to make access to the agent more immediate: instead of searching for an app every time they need to interact with the AI, the user carries this interface in their pocket or clipped as a keychain.

But it is still too early to evaluate how this experience will work in daily life. In the announcement, Meta did not detail the price or the full technical specifications of the Charm and stated that it would release more information later.

For now, the most important thing is not in the specs, but in the shift in interface: Meta is trying to transform an AI agent into something that stops being just software accessed on a screen and starts physically accompanying the user.

Behind the Charm is Muse, and it does more than just chat

To understand the Muse Charm, one must look at the agent behind it. Muse was introduced by Meta as a personal artificial intelligence agent capable of answering questions and performing tasks on behalf of the user. According to the company, it can handle activities such as sending emails, researching information, planning trips, and interacting with services used on a daily basis.

This capability is supported by the Muse Secure VM, a dedicated virtual machine that hosts the agent and has its own browser. It is in this environment that Muse can access authorized services and perform specific tasks, even while the user is doing other activities.

The difference compared to a traditional chatbot lies precisely in this transition from response to action. Instead of just explaining how to book a trip or organize a task, Muse's goal is to participate in the execution of that process, within the access limits defined by the user.

Therefore, the Charm should not be understood merely as a small voice assistant. It functions as yet another entry point for an agent that can also be accessed via phone, computer, web and, according to Meta's plans, through its AI glasses.

Why did Meta put an advanced AI inside a device that looks retro?

The design of the Muse Charm catches the eye precisely because it contrasts with the technology behind it. Instead of opting for a futuristic look, Meta presented its personal agent in a small device reminiscent of old Tamagotchis that can be carried as a keychain or accessory.

This choice comes at a time of renewed interest in technologies associated with the 1990s and 2000s. TechCrunch linked the Charm to the recent popularity of retro products and so-called bag charms, accessories used to personalize bags and backpacks, especially among younger consumers.

This does not mean that Meta has declared nostalgia as an official product strategy. However, the design helps bridge the gap between a complex technology and a familiar object. Instead of presenting the agent as just another tool inside an app, the Charm gives Muse a physical presence, with its own avatar and an interface created to accompany the user throughout the day.

In this sense, the Tamagotchi-like look may be less important for the nostalgia itself and more for what it represents: an attempt to turn artificial intelligence into something everyday, personal, and always close to the user.

Routine automation is the real product behind the character

The avatar and the design of the Muse Charm help make the device recognizable, but Muse's appeal lies less in its appearance and more in its ability to take over parts of the user's routine. Meta presents the agent as a system capable of executing tasks, organizing projects, and turning broader goals into action plans.

In practice, this includes activities such as sending emails, searching for flights, filling out forms, or performing actions in connected services. Some tasks can continue to run even after the user closes the app, and Muse prompts for attention again when it needs approval, such as before sending a message or completing a purchase.

The Charm adds another layer to this concept by reducing the distance between the person and the agent. While previously it was necessary to access Muse via phone or computer, a dedicated device allows this interaction to start more immediately throughout the day.

This convenience, however, depends on context. To help in an increasingly personalized way, the agent needs to understand preferences, tasks, and information related to the user's routine. And that is exactly where automation and privacy begin to be part of the same discussion.

The more personal the agent, the more data enters the relationship

For Muse to anticipate needs and execute tasks on behalf of the user, it requires access to information that goes beyond an isolated conversation. It is this combination of permissions, memory, and interaction history that makes the agent more personalized, but, on the other hand, also increases the amount of data exchanged.

Access begins with permissions

According to Meta, each person chooses which services they want to connect to Muse and what level of access will be granted. For an email service, for example, it is possible to allow only reading messages or also authorize sending. These permissions can be changed or revoked later. Credentials and authentication tokens are stored separately within the Muse Secure VM, and the company states that the agent does not directly view passwords or payment information.

Memory turns context into personalization

Muse also keeps information about what it considers relevant for each person. This allows it to recall details mentioned earlier and use them in future suggestions. Meta states that the user can view and edit files stored in the VM and ask the agent to forget specific information it has learned.

Some interactions can also help improve the models

Personalization is not the only purpose of this data. Meta reports that conversations, tool calls, and interactions between Muse and other agents, called trajectories, can be used to train new model checkpoints. According to the company, these interactions go through a process of removing relevant personal information before this use, and there is an option to opt-out of having the data used for training.

It is precisely at this point where convenience and privacy begin to intersect: the more context the agent receives to become useful, the more important it becomes to understand what happens to that information after the interaction.

Could the Pokémon GO case be repeating itself? The difference is in the data

Muse prompts an interesting comparison with another topic already discussed on CodeBlog: the Pokémon GO case.

For years, Pokémon GO transformed players' movements, images, and interactions with physical space into data that contributed to systems capable of understanding and mapping the real world. The product was a game, but part of what its users generated during the experience ended up creating value for a much larger technological infrastructure.

With Muse, the material used is different: it gathers conversations and logs of actions executed by the agent. These interactions can contribute to the training of models, under the conditions and controls described above.

This does not mean that Muse was created as a collection strategy similar to Pokémon GO. There is no basis to assert that. The parallel lies elsewhere: digital products can offer a useful experience while the interactions generated during their use help improve more advanced systems.

These logs can provide examples of how users phrase requests and how the agent executes tasks in real-world situations.

The Muse Charm might be less about a gadget and more about the next AI interface

It is still too early to know if the Muse Charm will be a mass-market product. Dedicated AI devices have already faced difficulties proving they offer something that smartphones, watches, or headphones cannot. Reuters highlighted this point by recalling the limited performance of products like the Rabbit R1 and noting analysts' assessments that the Charm will need to justify its existence beyond its initial appeal.

But perhaps the main signal is less about the device and more about the strategy behind it. At Meta Connect, Zuckerberg stated that the company has shifted its focus to devices geared toward what he calls "personal superintelligence." Muse is also expected to come to Meta's smart glasses, expanding the access points to the same agent.

In this scenario, the Charm serves as an experiment in how AI can transition from being just a service accessed inside an app to physically accompanying the user across different interfaces.

The design resembles a Tamagotchi, but the bet is different: to make the agent a constant presence in the daily routine. Whether this will create a new device category remains uncertain. What already seems clearer is that the fight for the next artificial intelligence interface has begun to move off the phone screen.