Meta put Muse in a pocket device before publishing the product details

01

WHAT ACTUALLY CHANGED

Meta revealed Muse Charm during its annual Connect event on September 23.

Meta's official Connect summary describes Charm as a pocket-sized device made for speaking with Muse and says more details will arrive later in 2026.

Reuters reported from the event that the device is approximately the size of a pocket watch and has a roughly two-inch, or five-centimeter, display.

Reuters says the display can be navigated with a thumb and centers a customizable Muse character.

The device has a built-in 5G modem, allowing access to Muse away from Wi-Fi, according to Reuters.

Charm cannot make telephone calls, despite having cellular connectivity.

Users can tap the device to reach Muse directly rather than opening an AI application on a smartphone.

Mark Zuckerberg said Meta had packed the Muse voice and avatar experience into a small device and was shifting its device work toward what he calls personal superintelligence.

Zuckerberg also said Muse would come to Meta's smart-glasses lineup, placing Charm inside a broader strategy rather than presenting it as a lone experiment.

Analysts quoted by Reuters said the object may appeal to younger users through personalization, fashion and nostalgia, while warning that dedicated AI handhelds have struggled to justify themselves beside phones, watches, earbuds and glasses.

The public reveal did not include a complete specification sheet, price, order page, release date or supported-market list.

Meta has not publicly identified Charm's processor, memory, storage, battery capacity, expected battery life, charging method, durability rating, repair options or security-support period.

Meta has not said whether Charm runs a language model locally. Reuters quoted an analyst discussing that as a possibility, not as a confirmed capability.

Meta has not published a complete list of Charm's microphones, cameras, motion sensors, biometric sensors or physical privacy controls.

The company has not explained which Charm interactions remain on the device, which move through the cellular network to Meta systems and which can involve outside services or people.

02

WHY THIS MATTERS

A dedicated device should solve a problem that the phone does not. Faster access can be useful, but one less tap is a thin reason to buy, charge and secure another computer.

The no-calls detail is clarifying. Charm uses a cellular network to reach an agent, not to replace the basic communication function of a phone.

Built-in 5G turns the device into an always-available network endpoint. The missing questions include carrier support, service fees, roaming, data use, lost-device controls and behavior when coverage disappears.

Voice access requires at least a microphone, but the reveal has not established the complete sensor package or whether a physical mute switch exists. A small object carried in public needs an unambiguous recording state.

A camera should not be assumed simply because the product has a screen or an avatar. If Meta later includes one, the company should state its field of view, activation rules, indicator design and retention behavior before sale.

Bystanders did not buy the device or accept its terms. They still need a visible, understandable signal when their words or images can enter an agent workflow.

Local processing changes the privacy, latency and offline story. Cloud processing changes the network, retention and account-risk story. Meta has not yet drawn that line for Charm.

A claim that a small language model might run locally came from an analyst quoted by Reuters. It should not be promoted into a product fact until Meta publishes the architecture.

Muse can act across email, travel, shopping and forms. A tap that begins a conversation may eventually lead to consequential actions, so the device must separate listening, drafting, approving and executing.

A tiny screen limits how much evidence and context can be shown before approval. That creates a human-factors question: can a person really inspect the destination, price, dates, account and cancellation terms before authorizing an action with one thumb?

The device is also an account-recovery problem. If it is lost or stolen, the owner needs a fast remote lock, revocation and activity review without depending on the missing gadget.

Meta says Muse runs on a secure virtual machine and gives users control over access. The Charm reveal does not yet explain how that software architecture maps onto the hardware, cellular connection and local credentials.

Reuters previously reported that human contractors completed some calls during Muse testing without that arrangement being disclosed to testers at first. Hardware disclosure should therefore include human handoffs, not only chip and battery details.

A friendly avatar can make a system feel social and trustworthy. It cannot communicate whether the agent is recording, waiting, sending data, asking a human for help or about to spend money unless the interface deliberately shows those states.

Earlier dedicated AI gadgets struggled when their capabilities, battery and reliability failed to beat the phone already in a customer's pocket. Charm will be judged against that practical baseline, not against the applause of a launch demo.

The social value could still be real. A small voice-first device may help people who dislike glasses, have difficulty navigating phone interfaces or want quick access while their hands are occupied. Those benefits need accessibility testing, not demographic guesswork.

A product that becomes personal by remembering a person also needs a memory panel, retention schedule and deletion path. Personalization without inspectable memory is just an invisible dossier with a cheerful face.

The most important specification may be the support lifetime. An agent gadget tied to cloud systems can lose core functions if models, subscriptions, networks or company priorities change.

03

WHERE IT COULD HELP

• Publish one product sheet that distinguishes confirmed launch features from planned capabilities and leaves no room for analyst speculation to become a specification.
• List every sensor, its purpose, when it activates, what indicator appears and whether a physical switch can disconnect it.
• Draw the data path for a simple voice question, a web task, an email, a purchase and a call that requires human help.
• Label each processing step as on-device, Meta cloud, outside service or human contractor, with the retention period for each copy.
• Provide a real hardware mute control whose state does not depend on software and remains obvious when the screen is asleep.
• Use distinct visual and audible states for idle, listening, recording, uploading, waiting for approval, acting and involving another person.
• Let bystanders recognize those states from several feet away without having to know the product or read a tiny icon.
• Require a second, explicit confirmation for payments, bookings, messages, account changes and actions that cannot be easily reversed.
• Move high-context approvals to a larger trusted screen when the two-inch display cannot show the destination, amount, dates and consequences clearly.
• Show the exact account and service Charm will use before an action, especially when one person has multiple email, shopping or travel profiles.
• Give users a plain-language memory dashboard where they can inspect, correct, export and delete what Muse has retained.
• Offer an offline mode that clearly states which features still work and prevents a network failure from masquerading as a completed task.
• Publish realistic battery tests for idle carry, frequent voice use, weak cellular coverage, hot weather and continuous avatar animation.
• Explain the cellular plan, supported carriers, international roaming, emergency limitations and what happens when a subscription ends.
• Provide a lost-device page with remote lock, credential revocation, recent activity and a way to recover access without the Charm.
• Commit to a minimum period for security updates, replacement batteries or repair, and continued core service.
• Test the device with blind and low-vision users, people with limited dexterity, Deaf and hard-of-hearing users and people with speech differences before treating voice-first access as universal.
• Run independent comparisons against the same Muse tasks on a phone, watch, earbuds and glasses, measuring setup time, completion rate, correction burden, privacy comprehension and battery cost.
• Report how often the agent needs a human contractor, how that handoff is disclosed and which data the contractor can see.
• Make the product earn its pocket by publishing three jobs it performs materially better than a phone, with the test method and failure rate attached.