We have Muse Charm at home.

Mark Zuckerberg’s Meta made AI look so cute by giving it a squishy face, and this holiday season, he wants to sell you a Muse Charm so you can carry your cheerful AI helper around like a Tamagotchi.

But I won’t buy one because we have Muse Charm at home. It’s only $50 and I’ve decided the learning experience was worth it.

My version doesn’t have cameras or a 5G cellular radio, but it actually talks to Muse and has cute avatars too! It doubles as a pocket watch and remote camera viewer, and can (theoretically) control my smart home.

Or at least it can be if you inject a lot of additional vibration code.

A few weeks ago, I paid $50 to get a 1.75-inch touchscreen hanging computer shipped directly from China, specifically the Waveshare ESP32-S3 1.75C. The push-to-talk AI development kit not only looks like a batch like Muse Charm, but is officially approved by Meta. It’s one of the company’s first suggestions if you want to create a “Muse Gadget,” a device that uses open source code and Meta tokens to connect directly to your Muse AI cloud.

When the silver metal pocket watch arrived, it started to look pretty useless. Setup was easy enough: I simply plugged it into my PC and pointed Claude to Meta’s GitHub, then paired it with the Muse phone app. But even after I put the Meta keys on the device, Muse originally thought I was sending it audio clips. transcribe instead of commands. After I clarified that, Muse’s responses were still slow and he didn’t respond out loud to me. It was hard to tell what it was saying, as it only displayed a few words on the screen at a time, with no place to read the entire messages unless I pulled out my phone with its own Muse app.

The Meta version also only came with a single Jolly avatar that mostly just sat there and looked cute, and wouldn’t automatically discover smart home devices like Amazon Echo or Google Home do. (This device didn’t really work anything useful after my initial installation.) And while Meta AI software engineers Anant Narayanan and Jeremy Payne include a neat sample app showing how Muse Gadgets can stream video from your phone’s camera, the lack of a remote shutter button made it not as useful for taking photos.

But a few Claude Code sessions filled almost all those holes. All I had to do was reconnect Waveshare to my PC and then ask Claude to reprogram every single thing I wanted.

When I complained to Claude that I couldn’t read or hear Muse’s answers to my questions properly, problems that others have also mentioned, Claude created a conversation history page so I could swipe right to read what he was saying, and I connected a British AI voice from ElevenLabs that speaks his answers out loud. (The voice was the least soul-sucking of the free ElevenLabs tier, my wife and I agree, even if it doesn’t match the cuddly vibe of the Muse avatars at all.)

Thanks to Claude, I can now press a button for my phone to take a photo from across the room with a high-quality live preview, and my phone’s front camera is no longer displayed upside down. Claude also made pretty short work of adding a low-quality camera preview over Bluetooth, so I don’t need to be connected to home Wi-Fi to get photos from creative angles.

I asked Claude to create a whole batch of different avatars like the ones Meta showed off during Meta Connect, and to add a Tamagotchi-like minigame where they need food, play, sleep, and a bath. When the game seemed too simple to hold attention, I asked him to add four different levels of hunger, boredom, and dirt. I decided that treats should make a crunchy mess and make the avatars messier when they eat them, and that too much play should tire out a pet. My wish was Claude’s order.

When I realized I was accidentally releasing the push-to-talk button too early and causing Muse to choke on nearly empty audio, vibration encoding fixed it minutes later by sending only audio over a second long. The only obvious lag I see now comes from the Meta servers themselves: sometimes they respond, sometimes they lag quite a bit.

It was so easy to let Claude reprogram the pendant that I asked him to create some features specifically for this story video, including a functional pocket watch face with three hands that actually tells time. (That took 15 minutes; now when I say the hardware looks like a pocket watch, you can see what I mean!) He even found, downloaded, trimmed, and uploaded video clips of Zuck presenting Muse Charm and installed a video player so he could play those clips on Waveshare.

I want you to understand that I have no coding experience, so I have no idea if any of this code is any good. It could have a fatal error. I also don’t trust Meta or their agent Muse my email or home network, particularly after hearing about how a reviewer’s home address was leaked. (I put it on guest Wi-Fi early on as a precaution.)

That means no In fact know how useful a muse amulet may or may not be for those who do know how to code and are willing to take a greater leap of faith. But after almost dropping a small, bulky silver pendant a couple of times, one that has no more Muse functionality than my phone (and quite a bit less without additional code), I think Muse itself is not the reason to buy a dedicated computing pocket watch.

The reason for having a dedicated piece of hardware is to do things that you not able do on the phone that’s already in your pocket, like encoding the entire device to do whatever you want. Which makes me wonder: will the real Muse Charm be as freely programmable, or would the type of people who would buy one be better off with a development kit? I guess we’ll see when it ships.