Making a phone call shouldn’t feel like learning a new piece of software. The HMD 2660 Flip Senior Edition responds to that familiar frustration with big buttons, a folding body, and an unexpected addition called Mary. This AI companion from inTouch brings conversation to a basic phone, letting older adults try the technology without trading their keypad for a touchscreen full of apps.
The hardware doesn’t stray far from the regular HMD 2660 Flip introduced in 2025. Open the clamshell, and there’s a 2.8-inch screen above a generously sized number pad, with a navigation control separating the two. Answering or ending a call can be tied to opening or closing the phone, giving those actions a physical beginning and end that a touchscreen doesn’t quite offer.
The outside gets its own smaller display, so checking the time doesn’t require unfolding the whole thing. Its dark body and rounded corners stick to the recognizable flip-phone formula rather than dressing the device up as medical equipment. That restraint matters here. The phone offers an alternative way to stay in touch, without announcing that its owner needs an entirely different category of gadget.
Mary adds a different kind of contact to the phone, one that’s available for a casual conversation around the clock. HMD says the companion speaks slowly, leaves room for replies, and remembers earlier conversations. A chat might involve history, a favorite memory, or a game. The interesting choice is delivering those interactions through speech, instead of asking someone to type their thoughts.
Of course, an AI feature can easily complicate a supposedly simple phone. Here, Mary appears in the main menu, and selecting her icon before pressing the call button starts the conversation. There’s no separate app to install or number to memorize. Activation still involves a website or QR code, though, so the simplest experience comes after someone has handled that initial setup.
The less glamorous details haven’t disappeared beneath the AI pitch. A removable 1,450mAh battery and USB-C charging keep the basics familiar, while HMD rates the body IP54 for protection against dust and splashes. A replaceable battery also means you can swap a worn component, rather than treating the whole phone as something that has reached retirement age.
The UK listing puts the phone at £79.99 (about $107), including a five-hour trial of Mary. Continuing the service costs £16.90 (about $23) monthly, with carrier charges potentially adding to that. The trial doesn’t automatically become a paid plan, but inTouch’s general terms say paid subscriptions renew automatically. It can be canceled, but ongoing conversations introduce another expense alongside ordinary phone service.
There’s an important distinction between making conversation easier to access and replacing human company. Mary is software, and her availability shouldn’t be confused with evidence that she reduces loneliness. The change here is how someone reaches that software. They can keep the familiar ritual of opening a phone and making a call, and still use their handset after the AI subscription ends.
Copying customer details from a message into a spreadsheet probably isn’t anyone’s idea of a good afternoon. Unfortunately, even a perfectly capable computer still leaves plenty of that work to you. Bvio’s Violoop is a plug-in AI operator designed to handle those repetitive steps, giving your existing computer an assistant without asking you to replace the machine you’re already comfortable using.
You can tell Violoop what you need in ordinary language, such as finding customer information in a WhatsApp conversation and entering it into a customer relationship management (CRM) system. That might sound like a small convenience, until you’ve spent an afternoon repeating the same routine. The aim is to let you explain the task instead of manually carrying every detail from one window to another.
To do that, Violoop interprets what’s on the screen and interacts with applications through keyboard and mouse controls. Its companion software supports Windows, macOS, and Linux, while the screen-based approach means it doesn’t need a dedicated integration for each application. You’re still working with familiar programs, only with assistance intended to bridge the gaps between them, and without the need to learn programming to give it instructions.
Of course, customer records are only part of the paperwork that follows a small business around. Invoice details need entering into spreadsheets, documents need organizing, and files rarely end up where they belong without some encouragement. Violoop is designed to help with that preparation, taking on the repetitive handling while leaving you to check the information and decide what to do with it.
Email offers another familiar example. Having something sensible to edit can be easier than staring at an empty reply box, especially when several messages need similar responses. Violoop can prepare replies as drafts for review, giving you a starting point while keeping the final wording in your hands. You can add the context that a customer relationship or a delicate conversation actually needs.
There’s also a reason this assistant takes up physical space on your desk. Its compact enclosure and angled touchscreen give ongoing tasks a place of their own, rather than burying everything in another software window. Inside, the tiny device packs a punch of up to 26 trillion operations per second (TOPS) of AI processing. That hardware is intended to support local screen understanding and other AI work on Violoop itself.
That local processing also supports a more personal approach to recurring work. Violoop keeps personal memory on the device, retaining preferences and context so you don’t have to explain the same habits every time. Local memory doesn’t mean every task stays offline, however. Depending on the selected mode, Violoop can also connect to cloud services or larger AI models when additional help is needed.
The most interesting control might be the button on top. Violoop separates AI processing from a dedicated security controller, with an approval step for sensitive actions. The physical confirmation button gives that pause a tangible form. Preparing information and committing to an action are different responsibilities, after all, and the design puts a deliberate checkpoint between what the AI proposes and what happens next.
You don’t have to stay beside it to keep an eye on progress, either. Phone access lets you check tasks, send further instructions, and approve sensitive actions while you’re away from the desk, provided the computer remains on and connected. That’s useful when you need to step out but still want to see how things are going, without returning to the keyboard for every update.
An organized spreadsheet won’t decide which project you should take on next, and a drafted reply doesn’t know everything about the person receiving it. Those parts still belong to you. Violoop’s role is to help get the surrounding work ready, using a device that keeps progress visible and approval close at hand. That leaves the final email waiting for something decidedly familiar: your own words.
Interchangeable batteries offer near-uninterrupted use
Familiar frame styling keeps the design approachable
Adjustable temples and nose pads help personalize the fit
In-view translation into 98 languages reduces the need to check a phone
CONS:
Swapping batteries require a 5-second reboot
Optional prescription lenses don't support hyperopia (farsightedness)
RATINGS:
AESTHETICS
ERGONOMICS
PERFORMANCE
SUSTAINABILITY / REPAIRABILITY
VALUE FOR MONEY
EDITOR'S QUOTE:
The INMO GO3 AI Glasses pair familiar eyewear styling with practical translation tools and the convenience of swappable batteries.
Looking down at a phone to translate a conversation can interrupt the exchange it’s supposed to help. Reading presentation notes creates a similar problem, pulling attention away from the people in front of you. The INMO GO3 AI Glasses bring those words into view instead, putting translation and presentation tools inside a familiar rectangular frame that doesn’t require a dramatic change in personal style.
That familiar shape is only part of the appeal, though. Adjustable contact points, discreet controls, and, most importantly, a replaceable battery give the GO3 a practical foundation beyond its display. Together, they suggest an approach to smart glasses that puts everyday wear on equal footing with software features. How convincingly the GO3 brings those priorities together comes down to a few details, starting with the frame itself.
The INMO GO3’s rectangular lenses and softened corners give it the outline of conventional eyewear, although the frame has more substance than a pair of reading glasses. That’s a reasonable visual compromise, of course, given what it packs inside. The thicker brow provides room for the display hardware while giving the front a confident shape, rather than leaving its components looking like something awkwardly attached afterward.
The dark matte finish does much of the visual work. It gives the broad surfaces a quieter presence than glossy plastic would, while the small red accents around the hinges add a bit of personality. Thankfully, the color doesn’t spread across the frame. There’s enough contrast to make the details interesting without making the glasses look like equipment borrowed from a gaming setup.
The GO3 utilizes aluminum alloy and polyamide for the construction with a CNC-finished, fingerprint-resistant surface. It’s not an entirely metal object, which makes sense when you consider that the electronics also add their own weight to the equation. The frame and temples share a restrained finish instead of competing for attention through contrasting textures.
The slim temples measure around 8mm, keeping the side profile from becoming an oversized box. The curved ends retain the familiar outline of glasses, even with additional hardware inside. It isn’t dainty eyewear, admittedly, but the shape avoids the visor treatment that would make a business lunch feel like a technology demonstration.
The camera is integrated into the front rather than perched above it, and the controls follow the temples. These details help the INMO GO3 read as a somewhat fashionable object instead of gadgets sharing a frame. A supplied camera cover also provides a way to obscure the lens when recording isn’t appropriate, especially given the current conversations around smart glasses.
Ergonomics
At approximately 58 grams, the INMO GO3 has the weight of electronics to accommodate as well as the frame itself. Comfort depends on more than the number on the specification sheet. The relationship between the nose pads, temples, and ears matters as much. There’s a focus on those contact points, rather than relying on a lightweight label to explain away the fitting process.
Air-cushion nose pads support the front, while the temples offer 15 degrees of adaptive movement, at least on paper. That flexibility gives the frame room to accommodate differences in head shape. Naturally, there will be cases when the head shape just won’t fit, but INMO has tried to accommodate as many head sizes as possible.
Given we’re talking about glasses, prescription support is another important piece of that everyday equation. The GO3 uses an insert system, with INMO’s current lenses supporting nearsightedness and astigmatism, though not farsightedness. That won’t cover every prescription, but it gives supported wearers a way to combine vision correction and the display.
Operating the glasses doesn’t have to depend on talking to them. Temple touch controls and physical power and GO buttons provide alternatives, while the companion app handles setup and settings. When you’re around people or somewhere that requires a degree of silence, a discreet gesture is often preferable to a spoken command, and familiar controls help make the wearable format a little less foreign.
Performance
Two-Way Dialogue Translation – Read translations directly through the glasses, while the other person sees or hears your translated response on the phone.
The INMO GO3’s binocular green Micro-LED display uses optical waveguides, with a 640×480 resolution and a 30-degree field of view. Adjustable brightness is listed up to 1,500 nits, which is plenty even for the bright outdoors. That said, those specs clearly mark it as something more for text and guidance rather than a miniature cinema. Green lettering has a slightly retro character, but a translation or the next line of a speech doesn’t need a full-color backdrop to be useful.
Translation, of course, is the primary use case for these glasses, and the numbers are admittedly impressive. INMO lists 78 source languages and 98 target languages online, with subtitles appearing in view rather than on a phone held between speakers. Offline support covers nine languages on Android and 13 on iOS, and future updates promise even more. The phone still needs to stay paired to a phone, of course, but a missing internet connection needn’t end every supported conversation.
Built-In Teleprompter – Follow your script directly through the glasses.
For presentations, the teleprompter moves the script from a desk into the speaker’s field of view, allowing for a more natural-looking stage presence. There’s automatic voice-following advancement in 17 languages (including English, Mandarin, Cantonese, Japanese, Korean, German, French, Thai, Russian, Spanish, Portuguese, Polish, Arabic, Indonesian, Vietnamese, Dutch, and Italian), alongside manual scrolling from the temple. Other languages will be displayed, depending on the document uploaded by the user, but the pages for unsupported languages can’t be automatically turned by voice. Very few will be sitting close enough to realize you’re not exactly looking at them, but at least you won’t be spending half of the presentation staring down at your notes.
Photo Translation – Point the camera at written text and read its translation directly through the glasses.
The GO3 also offers features better used in meetings, such as recording, transcription, and AI summaries for later review. Walking and cycling directions, an AI assistant powered by ChatGPT and Gemini, and photo translation extend the same idea beyond work scenarios: put a useful answer nearby without making a phone screen the center of every activity.
Admittedly, many AI smart glasses offer similar features, but the INMO GO3’s killer feature is its interchangeable battery system. One 270mAh battery is installed in the glasses, with one spare included. Continuous use yields around 3 hours on a full charge, which is probably more than enough for most use cases. The 1,300mAh battery charging case offers three to four more charges, so you practically get a full day’s worth in total.
Where it gets interesting is how you can swap out the dead battery on the tip of the GO3 with the spare in the charging case. It’s quick and convenient, but not entirely seamless. The glasses will shut down when you pull out the battery, but plugging a fresh one in boots it up in just a few seconds. It does mean you have to briefly stop what you’re doing or make sure you’re not in the middle of something critical, but it’s better than waiting for minutes for the glasses to charge up.
Sustainability
The replaceable battery has a longer-term benefit beyond getting through a busy afternoon. An aging power cell doesn’t automatically have to take the entire pair of glasses out of service. INMO lists a replacement battery at $49, making that component a separate ownership expense. That’s a useful piece of product design: the part expected to wear with repeated charging doesn’t have to remain permanently sealed inside.
That doesn’t establish a complete sustainability story, and there’s no need to pretend otherwise. The reviewed specifications don’t establish recycled material content, comprehensive component repairability, or an end-of-life program. The more defensible advantage is the ability to keep the same frame in service when its battery needs replacing. Thoughtful care still matters, particularly for the lenses and hinges, but accessible power replacement is a concrete place to start.
Value
At $699, currently discounted to $509, the INMO GO3 asks for a commitment beyond casual curiosity. The standard package includes the glasses, installed battery, single spare, and battery charging case. That’s a coherent starting point, with the power system already accounted for. The value comes from using the display repeatedly, rather than paying for a collection of functions that remain untouched save for a few presentations or travels abroad.
The useful comparison is between priorities, rather than a contest over the longest feature list. Camera-and-audio glasses answer a different need from eyewear that puts text in view, while entertainment-focused displays serve another purpose again. For someone moving between multilingual conversations and presentations, the GO3’s combination is easier to justify. That it can also be your travel companion and AI assistant brings added value to the table, especially when you don’t look like a walking technology advertisement.
Verdict
The INMO GO3’s most appealing quality is how its design keeps returning to recognizable habits. Wear glasses, follow a conversation, glance at a script, and change a battery when necessary. The technology adds steps of its own, of course, but the physical form doesn’t ask for an entirely new routine. Its familiar outline and adaptable contact points give the translation and presentation tools a sensible place to live.
Travelers, presenters, and people working across languages have the clearest reasons to appreciate that combination. A paired phone remains part of most workflows, and battery changes belong between tasks, but neither qualification takes away the appeal of information placed closer to the conversation. There’s something refreshingly practical about glasses that leave room for ordinary interactions, even when a little green text is helping them along.
Getting stuck on homework doesn’t always call for someone to hand over the answer. Sometimes a small nudge is enough to get the pencil moving again. Nico is an AI learning companion concept built around that idea, putting a helpful character inside a handheld device with a joystick. Children would steer him toward a question and receive a hint rather than the solution.
The hardware borrows the familiar arrangement of a handheld game console, with a large screen above an orange joystick and two matching action buttons. A dark gray body and rounded corners keep the silhouette simple, while the contrasting controls make the places to press immediately obvious. There’s something reassuringly straightforward about having a proper button instead of another tiny icon to tap.
A side button would open the camera view, bringing Nico into the child’s surroundings through the display. The character could then move across that view under joystick control, stopping beside whichever exercise needs attention. A microphone and speaker are intended to let children ask questions aloud and hear responses, so getting help wouldn’t require typing a carefully worded request on a little screen.
With a math worksheet, for example, Nico could point out that multiplication comes before subtraction without filling in the blank. The child would still have to work through the calculation, ideally returning to the paper with a better idea of what to try next. That distinction matters, because finishing an assignment and understanding how to do it aren’t necessarily the same achievement.
Nico himself is a small white character with a rounded rectangular head, dark eyes, and orange accents that echo the hardware. Keeping that personality on the display would let the same device become a homework helper or a virtual home. The wrist strap and USB-C port, meanwhile, ground the friendly face in the ordinary business of carrying and charging yet another electronic companion.
Of course, Nico wouldn’t disappear when the homework is done. The concept includes mini-games, a room to explore, and a virtual house where children could cook with the character or look after him. That gives the joystick more to do, but it also introduces a familiar complication: the device helping with fractions would be offering somewhere else to spend the afternoon.
The camera and microphone raise questions that the presentation doesn’t answer. There’s no stated approach to storing children’s images or conversations, no explanation of parental controls, and no identified AI model behind the advice. These are unresolved parts of the concept, not evidence of a privacy breach. But they would need attention before a friendly face earned a place beside a child’s schoolwork.
Nico remains a design proposal, with learning benefits that haven’t been established through published independent testing with children. Its joystick does make a specific suggestion about how children might ask for help, though: move toward the problem instead of waiting for an answer to appear. The important test would come afterward, when the device goes down and the pencil starts moving again.
Keeping in touch shouldn’t require learning another app, but small icons and unfamiliar menus can turn a simple family message into an unwanted chore. Ato approaches that problem with a screen-free AI voice assistant for older adults, packaged more like a tabletop radio than a computer. Designed by Boostep for Ato, the new version has progressed through physical prototypes, with customer deliveries scheduled for the future.
The rounded cream housing and woven front borrow their character from familiar domestic electronics. There’s no glossy display competing with the room, just a textured face and a bright orange button. The resemblance is intentional, drawing on radios, record players, and televisions. It gives the assistant something recognizable to wear without pretending that everyone over a certain age wants the same thing.
That orange button gives the voice-first interface a physical starting point. Press it and begin talking, rather than remembering which icon opens a conversation. A large volume knob projects from the side, with small dimples around its edge. Turning something you can locate by touch seems a sensible alternative to searching for tiny volume symbols on a smooth and otherwise featureless surface.
The family side of the arrangement still uses a smartphone app. Relatives can set reminders and send messages, while the person at home interacts through speech. An appointment reminder or a note about Sunday lunch could arrive without making the recipient navigate the same interface. The complexity hasn’t disappeared altogether; it’s been placed where family members can manage it more conveniently.
Of course, Ato isn’t intended to wait silently for instructions all day. The company describes proactive check-ins and conversation, alongside radio and spoken information. That gives the device reasons to stay near a favorite chair instead of becoming something used only for reminders. Companionship is the intention, though, not evidence that an AI conversation can replace time spent with other people.
At approximately 5.5 inches tall, the enclosure is small enough to sit beside a chair rather than dominate a shelf. USB-C supplies power, and there’s no battery inside, so its place on a shelf still needs an outlet nearby. Wi-Fi is required too. You can move it between rooms, but power and network access determine where it can actually keep you company.
A familiar exterior doesn’t settle questions about listening and privacy, however. Ato says it doesn’t retain audio recordings or let relatives read conversation transcripts, although it stores selected personal facts for future conversations. That matters when conversation is the product’s main purpose. The company also states that Ato isn’t a medical device and cannot contact emergency services, so it isn’t a substitute for dedicated emergency equipment.
The device is listed at $179 plus shipping, with deliveries scheduled for February 2027. Basic messaging, reminders, and radio remain free, while unlimited conversations require an optional membership listed at $249 annually. The upfront price therefore isn’t the whole decision. Someone expecting long daily chats would need to budget for the service, not just find a place for the speaker.
Compact, minimalist design with automotive-inspired detailing
Effective cooling system for sustained heavy workloads
Impressive performance for local AI use, graphics work, and light gaming
Great price value for its size and features
CONS:
Fan is clearly audible during high-performance tasks
USB-C ports are all at the back
RATINGS:
AESTHETICS
ERGONOMICS
PERFORMANCE
SUSTAINABILITY / REPAIRABILITY
VALUE FOR MONEY
EDITOR'S QUOTE:
The VZMORE AX9 Max Mini AI PC pairs understated design with practical local AI performance and cooling that keeps demanding work moving.
A desktop computer shouldn’t have to dominate the desk it sits on, particularly when that space also holds a drawing tablet, reference books, and the occasional coffee cup. The VZMORE AX9 Max takes that idea in a promising direction, pairing an understated enclosure with hardware capable of running local AI models and tackling extended workloads. Its vapor-chamber cooling gives this small machine an appropriately serious engineering focus.
Of course, fitting ambitious hardware into a compact box doesn’t automatically make a great desktop. Our review configuration combines AMD’s Ryzen AI 9 HX 470 processor with 32GB of RAM and a 1TB SSD. After testing local models, extended heavy tasks, display connections, and Fortnite, the more interesting story is how VZMORE balances those capabilities with everyday practicality, including a fan that makes its presence known.
There’s a familiar language to the AX9 Max’s silver enclosure, rounded corners, and square footprint. Comparisons with the Mac mini are understandable, although VZMORE’s interpretation is boxier and taller, giving it a slightly more substantial presence. It remains compact enough to share a desk comfortably, without adopting the angular styling that often makes performance-focused hardware look as though it belongs beside a gaming chair.
The aluminum chassis and metallic finish provide the visual restraint, while the shield-shaped badge on top supplies a little personality. There’s something faintly automotive about it, like an emblem borrowed from a luxury car rather than a computer. It’s a playful detail on an otherwise straightforward design, and it keeps the AX9 Max from looking completely anonymous among the many silver boxes competing for desk space.
That restraint also makes the design easier to live with beyond a dedicated office. It wouldn’t look particularly out of place beside a living-room display or on a shared worktable. The darker base and ventilation openings supply some visual separation from the metal body, but they don’t turn the cooling hardware into decoration. Most of the personality stays on top, leaving the rest pleasantly businesslike.
Ergonomics
Measuring 5.31 x 5.19 x 2.27 inches and weighing approximately 1.71 pounds, the AX9 Max is easy to relocate between workspaces. The included VESA mount and screws also let it disappear behind a compatible monitor. That’s a kind of flexibility that you’ll want to have when the desk serves several purposes, although the external power adapter still needs a home. Small computers don’t magically make their cables disappear, much as we’d like them to.
The front panel is refreshingly generous with four USB-A ports, a headset jack, and a power button. A side-mounted SD card slot adds another practical convenience for a photography desk. USB-C, however, stays around the back, so regularly connecting a portable drive or another USB-C accessory involves a little reaching. It’s a workable arrangement, though a front USB-C port would have made those frequent swaps easier.
Those rear connections include two 40Gbps USB4 ports and dual 2.5GbE networking, giving a compact workstation useful flexibility for storage and wired connections. Placement still deserves some thought, especially behind a monitor. The bottom intake and rear exhaust need breathing room, so this is a machine to give a clear, firm surface rather than tuck into a fabric-lined corner just because it happens to fit.
Performance
Inside, the Ryzen AI 9 HX 470 brings 12 CPU cores, 24 threads, and boost speeds up to 5.2 GHz. Its integrated Radeon 890M has 16 compute units, while our configuration pairs that hardware with DDR5 memory and PCIe 4.0 storage. These are substantial ingredients for a compact desktop, but their usefulness depends on how effectively the enclosure handles the heat generated once work gets demanding.
That’s where the vapor chamber earns its attention. Inside this sealed heat spreader, a working fluid evaporates near the hot area, carries heat across the chamber, and condenses in cooler regions. The process helps distribute concentrated heat across the cooling assembly, where fins and a fan can release it. It doesn’t eliminate the need for active cooling, but it gives that cooling system a more useful surface to work with.
During our extended heavy tasks, the AX9 Max maintained performance without an observed sustained slowdown. The fan was clearly audible, although we wouldn’t describe it as exceptionally loud. That’s a trade-off worth understanding rather than disguising: this machine lets you hear some of the work going into keeping it running. For longer processing sessions, that steady output is more valuable to us than pursuing silence at any cost.
For a creative desk, that balance makes the AX9 Max an interesting candidate for editing footage, preparing graphics, or CPU-based rendering. We haven’t measured a specific video export or rendering project here, so there’s no promise that every creative application will behave alike. The appeal is a compact platform with useful processing headroom; demanding GPU effects and large 3D scenes still need expectations appropriate to integrated graphics.
Local AI gave us something more tangible to explore. With up to 24GB of system memory allocated to graphics, a 27-billion-parameter Qwen model at Q4 quantization produced about 16 tokens per second in our setup. Quantization reduces the model’s memory requirements by storing its weights at lower precision. The result makes experimenting with a larger local language model practical, without having to accommodate a separate graphics card.
A smaller nine-billion-parameter Qwen model reached about 22 tokens per second at Q4, with multi-token prediction enabled, 16GB allocated to graphics, and a configured 32K-token context window. These are setup-specific results, not a guarantee for every model. That allocated memory is shared with the rest of the computer, too, so leaving room for other applications remains part of the balancing act when building a local AI workspace.
Our display testing revealed a useful distinction between the available connections. The 5K and 6K monitors worked properly over USB-C DisplayPort, giving us the high-resolution workspace we’d wanted. HDMI was less cooperative with those setups, producing flickering before the image disappeared. We haven’t established the cause, so it wouldn’t be fair to declare a universal HDMI problem. USB-C was simply the reliable option with our tested displays.
Gaming fits into the picture as an enjoyable extra. Fortnite ran in the neighborhood of 80 fps at 1080p in our testing, and 1440p also delivered a very playable experience. At 4K, however, frame rates sometimes dropped into single digits. Settings matter, and these results shouldn’t become a blanket performance promise. The practical takeaway is to enjoy its gaming ability at sensible resolutions rather than chase maximum pixel counts.
Sustainability
The metal enclosure is a welcome starting point for durability, although plastic remains part of the construction, particularly around the airflow base. We don’t have verified recycled-content figures or a life-cycle assessment to attach to it. Its more convincing sustainability argument concerns keeping the computer useful, rather than treating a metallic finish as an environmental credential. That’s a less glamorous story, but a more meaningful one.
Replaceable SO-DIMM memory and two internal SSD slots give the AX9 Max room to grow. The storage slots use different sizes, M.2 2280 and M.2 2230, so an upgrade needs the appropriate drive. That flexibility can extend usefulness as projects expand. VZMORE also provides a three-year limited hardware warranty for official-site purchases, although upgradeable components and warranty coverage don’t establish that every part is equally repairable.
Value
At $1,499 for the 32GB RAM and 1TB SSD configuration, the AX9 Max belongs in the premium mini PC conversation. Its price makes sense when the cooling system, local AI capability, connectivity, and compact design all serve your work. A cheaper office computer or a larger graphics-focused desktop answers a different brief. Here, the attraction is having several useful capabilities gathered into one fairly unobtrusive object.
Discounts, when they’re available, make the package more appealing, but the full-price argument rests on what you need. I’d consider it worthwhile for a compact local AI and content-creation setup, with the creative-work qualifications already noted. Silence-first buyers have another calculation. VZMORE lists 45W, 54W, and 65W profiles, giving users alternatives to selecting the most aggressive performance setting for every task.
Verdict
The VZMORE AX9 Max’s most appealing quality is how little its appearance tells you about the work it can attempt. Under that restrained exterior sits a computer that ran substantial local models and kept going through our extended heavy tasks. The cooling system is central to that experience, even when it’s audible. There’s a coherent relationship between the small enclosure, the airflow design, and the machine’s practical ambitions.
USB-C proved the better partner for our high-resolution monitors, and gaming made more sense below 4K. Those boundaries leave plenty of room for a capable everyday workstation, particularly if local AI is part of the plan. I can see the appeal of keeping a machine like this beside a drawing tablet and a good display. The desk stays available for the work, rather than being surrendered to the computer.
Remember those Radio Shack project kits with the spring terminals, where you bent wires into a working radio and felt like Tony Stark? Elecrow’s CrowPi 3 is that feeling, updated for a world where the thing kids want to build is a chatbot. It is a Raspberry Pi 5 learning station with 41 built-in sensors and modules, a camera, a microphone, and a 4.3-inch touchscreen. The courses run from blinking an LED to face recognition and voice control, plus plugging in LLMs like ChatGPT, Gemini, and DeepSeek. Kids already talk to AI all day, so Elecrow figured they should get to build one.
The CrowPi 3 ships in a bumpered gray briefcase with a frosted magnetic lid, a fold-out handle, and an orange pull tab, which makes it the friendliest cyberdeck I have seen. For the uninitiated, that is a DIY computer built into some rugged container, usually with a Raspberry Pi for a brain and a lot of sci-fi attitude. Most of them cost their builders a few weekends of 3D printing and swearing, while this one arrives finished. The faceplate is silkscreened with labels and pin numbers, so the board doubles as its own wiring diagram. A middle schooler can legally carry the whole thing onto the bus, and I did not own anything this cool at that age, so I am only slightly bitter.
Every module gets its own labeled cell, so the ultrasonic sensor, buzzer, tilt switch, RFID reader, and flame sensor sit in plain view instead of hiding behind a lid. That layout is the best teaching decision in the box, since a kid can point at the part they just coded and watch it react. Two breadboards on the left handle custom circuits, and a Pico, Arduino Nano, or micro:bit can join the Pi 5 on the same board. The unit measures 11.2 by 7.3 by 1.5 inches (285 by 185 by 38mm) and weighs 2.43 pounds (1,100g), about a hardcover textbook. Elecrow calls the design open source, with resources on GitHub.
The touchscreen is the one part of the faceplate that moves, folding flat or flipping forward for a better angle. A 2MP camera sits on a small bar above it like a laptop webcam, which is exactly where a face-recognition lesson wants it. At 800 by 480 pixels and 300 nits, the screen suits status readouts and debug info, and the It’s FOSS reviewer found the custom learning interface doesn’t display properly at that size. The full-size HDMI port solves that, so a bigger monitor can take over once the real coding starts. The RGB cooling fan got a gripe too, loud for its size and always spinning, though the same reviewer cheerfully called silencing it a free coding exercise.
The AI curriculum covers OpenCV (an open-source computer vision library) image processing, object tracking, face recognition, QR codes, pedestrian and car detection, and speech recognition. Elecrow says there are 180+ lessons in total, which is a lot of homework disguised as fun. The LLM lessons connect the board to ChatGPT, Gemini, and DeepSeek, and since a Pi 5 can’t host models that big, expect those to lean on online services. Elecrow’s project ideas are pleasantly small, like a voice-activated night light or a piggy bank that recognizes coins and plays a sound for each deposit. There is also a Minecraft course where tapping an NFC card spawns TNT, which is how you sneak Python into a kid’s weekend.
Raspberry Pi’s own 500+ answers a simpler question: what if the whole computer lived inside a keyboard? For $200 you get 16GB of RAM, mechanical keys, and a 256GB SSD, which makes a fine desktop and a poor science fair. It has no joystick, no ultrasonic sensor, and no flame detector, which are the things a kid actually wants to poke. The closest direct rival is the Joy-Pi Advanced, a Pi 4 kit that TechRadar put above $470 while comparing it to the CrowPi 3’s Kickstarter pricing. The CrowPi 3 is the one that sells the poking.
The original CrowPi hit Kickstarter in 2018 with a Pi 3B+ and a carry case, and the CrowPi 2 later fit in over 22 sensors and modules. The CrowPi 3 nearly doubles that count and adds the AI courses, plus Pico and micro:bit support that the earlier kits never had. Makers who build briefcase cyberdecks from scratch will get the appeal immediately, since this one skips the modeling and wiring marathon. The sensors come pre-wired and the lessons come already written. That is a fair trade when the person at the desk is eight.
Elecrow sells two kits, and the gap between them is a Raspberry Pi. CNX Software’s December 2025 roundup had the Basic Kit at $229 on Elecrow’s store and the Advanced Kit at $439, with the Advanced Kit listed at $599 on Amazon, so check that listing for today’s price. The Basic Kit expects you to bring a Pi 5, while the Advanced Kit adds a 16GB Pi 5, a wireless keyboard and mouse, two 18650 batteries, and two game controllers. Kids will spend the next decade chatting with AI, and the Radio Shack kids turned out fine because someone let them open the box. Somebody will build a piggy bank that judges your coins, and honestly, I would back that.
It’s 2026, and there’s a fair chance any kid born in the next 10 years will see more server racks than they will actual bears. An incredibly twisted reality, but a reality nonetheless… so much so that Rajeev Basu decided to poke fun at this predicament in a way that only artists with a dark sense of humor know how. Meet Bezzy, a stuffed plushie that’s actually a server rack, not a teddy bear. Squeeze it and it emits an incredibly annoying hum, captured from actual data centers around USA.
Bezzy would be cute if it weren’t so pathetic. The toy’s thinly veiled satire on the fact that our AI CEO oligarchs want us to think of their world-ending water-sucking creations as anthropomorphic harmless beings. Meta unveiled Muse, a seemingly cute avatar that hides the fact that it’s selling your data to governments and marketers. OpenAI unveiled Dots, an adorable set of AI assistants that definitely don’t want you to think about the fact that the company’s AI bots have been routinely hacking into organizations, governments, etc. To that extent, Bezzy might be too effective in making AI seem cute. That’s until you hug the thing.
Take a close look at Bezzy and you start appreciating how much thought went into making him look so deliberately unstudied. The gray sherpa fleece, the floppy limbs, the googly eyes, and that little smile are all pulled straight from the visual grammar of the charity adoption plushie, specifically the kind WWF used to send you after you donated to save a panda. His torso, though, is a different story: a black textured panel with rows of blue LED dots running down it, server-rack iconography dressed up in fur. All 20 were handmade in the US, and the design intentionally avoids looking polished or new, going instead for that lived-in quality of a toy you might find at the bottom of a thrift store bin. The yellow tag dangling from his chest, screaming “HUG ME FOR EAR PIERCING FOREVER HUM,” is doing most of the heavy lifting.
The toy is the physical centerpiece of a bigger satirical project called Adopt A Data Center, which Basura designed to look exactly like a charity adoption portal. Hit “adopt now” and instead of a certificate, you get a page of citations and journalism covering data center noise complaints, water consumption, land use, and the tax incentives that make communities such convenient targets for developers. Stickers in the campaign read “Open your heart. And your power grid.” The tagline for the whole thing is “All they ask for is everything,” which is honestly one of the better lines written about the AI era.
The duo behind the project is a pretty inspired pairing for something this arch. Rajeev Basu runs Basura, a New York studio with a resume that includes stints at Wieden+Kennedy, Spotify, and Maximum Effort, where he wrote for Ryan Reynolds and Rob McElhenney. His previous projects have the same DNA as Bezzy: the Hinge Phonebook was a phone-shaped book filled with ideas for getting off your phone that sold out in 24 hours, and the Yerba Madre Dirt Shoes literally crumbled apart mid-walk to spread wildflower seeds. His collaborator is Alan Wilkis, who makes music as Big Data, a project he started in 2012 explicitly to comment on how the internet was mining and monetizing personal data. Bezzy launches alongside Big Data 4.0, Wilkis’s first album in seven years, and a new single called “Rabbit Hole,” which honestly tracks.
The hum inside Bezzy wasn’t recorded in a studio. It was captured by a woman living next to a data center in Virginia and handed over during Basura’s research process, part of six months of conversations with people across the US who live beside these facilities. Northern Virginia alone hosts over 500 data centers handling as much as 70% of the world’s internet traffic, and residents there have been filing noise and health complaints for years with very little relief. A 2026 UN University report found that data centers could consume 945 terawatt-hours of electricity annually by 2030, while AI-related water use could equal the domestic needs of 1.3 billion people. Bezzy, all 20 of them, at $99 each, won’t fix any of that. But he’s a pretty hard thing to ignore while he’s screaming.
I know you will relate to one of these telltale moments. A meeting ends, someone gives you three useful details, and by the time you reach your desk, only one remains. The same thing happens after an interview, a doctor’s appointment, a lecture, or when someone rattles off directions while you’re already trying to keep up. We like to think we remember conversations better than we actually do.
Mobvoi’s TicNote Watch is designed to cover that very human failure. Instead of asking you to open an app, find a recorder, or reach for your phone, it gives the watch a dedicated physical recording button. Press and hold it to start recording. The recording can happen without a phone and even works offline. To me, the idea seems pretty interesting for capturing my team’s briefing.
Most smartwatches are defined by everything they can do. Notifications, workouts, calls, apps, payments, music, health tracking – the list keeps getting longer. TicNote Watch is different though, as its most important interaction is deliberately narrow. Just press a button and remember what just happened. That makes the hardware control more interesting than it might initially seem. There is something satisfying about a product that behaves more like a lighter, a film camera’s shutter, or a staple gun than another miniature touchscreen computer. The action is physical and doesn’t require you to navigate an interface while somebody is still talking.
In a world where dedicated controls have increasingly disappeared behind glass, putting one back on the wrist becomes a design statement. This smartwatch isn’t trying to make every interaction smarter; rather, it aims to make one interaction nearly impossible to mess up. The double-press gesture adds a smaller version of the same idea, letting users capture a quick thought of up to 60 seconds.
Simple input, surprisingly busy output
What happens after the button press is where the AI kicks in. TicNote Watch can transcribe conversations on the wrist, with live text supporting 17 languages, and users can tap to highlight important statements while a conversation is happening. Once the recording ends, the system can generate an AI summary and mind map. The intended audience for the wearable can be a journalist leaving an interview, a recruiter moving between candidates, a consultant bouncing between meetings, or a graduate student sitting through a campus lecture. For someone spending much of the day listening rather than typing, removing the act of note-taking can be genuinely useful.
There is, however, an obvious social wrinkle. A recorder that lives on your wrist is easy to reach for, but that doesn’t mean everyone around you will be comfortable being recorded. Consent and workplace etiquette matter, particularly in meetings or shared spaces. Mobvoi also claims more than 24 hours of continuous recording and up to 200 hours of local storage, so this isn’t unlimited memory in literal terms. At $249, with a limited-time introductory price of $199, TicNote Watch isn’t really competing on how many smartwatch features it can squeeze onto your wrist. It is built around a premise that doesn’t try to replace your phone, your notebook, and your entire day.
AI has plenty of serious questions to answer, from copyright and environmental costs to privacy. There’s also a more everyday hurdle: for many people, AI feels impersonal. An assistant might be useful for a quick task, but it doesn’t necessarily feel like something you’d want to carry around. Meta’s upcoming Muse Charm takes a different approach. It puts an AI agent in a tiny device and gives it a character you can get to know.
Muse Charm is a keychain-sized handheld device built for Muse, Meta’s AI agent. Rather than unlocking your phone and opening an app, you’re meant to reach for the Charm when you want to talk to Muse. You can still use the agent on a smartphone, so the device isn’t your only way in. Its pitch is simpler access: help that’s close at hand when a question or task comes up.
Designer: Meta
The Charm’s rounded-square, or “squircle,” shape gives it something of a Tamagotchi feel. In Meta’s released image, a dark screen and rounded casing frame a small beige character, while a cord hangs from the device. It looks more like a playful accessory than a miniature smartphone. That distinction matters: Meta isn’t just shrinking another screen. It’s trying to make an AI assistant feel approachable enough to bring along throughout the day.
The character is a fuzzy-looking on-screen mascot, not a plush covering for the device. Users can give it a name, adding a personal touch to an assistant that might otherwise appear as text or an abstract icon. Meta describes the Charm as a way to connect with Muse through voice, touch, and a camera. Together, the character and the small form factor turn a practical tool into something that resembles a digital companion, at least in appearance.
That cuteness is doing more than making the product photogenic. It gives Muse a recognizable identity and offers a gentler introduction to technology that some people find unsettling. Still, a friendly face is a design choice, not a solution to concerns about AI. Whether people come to trust Muse will depend less on how charming its avatar looks than on what the agent does with the access they give it.
Muse is intended to do practical work. The agent can read emails, book classes, draft documents, and make purchases, though Amazon has blocked purchases through it. Those are capabilities of Muse the agent; they shouldn’t be mistaken for a promise that every task can be completed entirely on the Charm without using another device. The benefit for users would be getting help with small jobs as they arise, rather than saving them all for a later phone session.
Picture remembering an errand while you’re out, or wanting to ask a question without opening your phone and getting pulled into notifications. That’s the problem Charm is designed to address. It could also appeal to someone trying to spend less time looking at a phone, although it’s too early to say whether owning another screen would actually reduce screen time. Its usefulness will come down to how smoothly those quick interactions work in real life.
Convenience has a trade-off. To work across email, shopping, and other services, Muse may need access to accounts and personal information. Readers will want to know what permissions they’re granting and how much control they retain over the agent’s actions. The Charm’s camera also makes privacy an important question to explore, rather than something its cheerful design can answer.
For now, Muse Charm remains a forthcoming product. Meta was still finalizing its materials and components when it was announced, had not provided a price, and was aiming to ship in time for the 2026 holidays. That leaves the biggest question open: will people find it easier to use than the phone they already carry? If Meta gets that right, the cute mascot may draw people in. The everyday usefulness of Muse will give them a reason to keep it close.
