A small touchscreen pendant can do more than answer voice prompts. With help from Claude Code, one owner turned the $50 Waveshare ESP32-S3 1.75C into a personalized Muse AI gadget with spoken replies, a virtual pet, and phone camera controls. The experiment suggests that the appeal of AI hardware may come as much from what people can build as from the assistant itself.
Starting with a basic Muse device
The Waveshare device is a 1.75-inch touchscreen computer designed for push-to-talk AI use. It is also among Meta’s suggested options for making a “Muse Gadget,” which connects to the Muse AI cloud using Meta’s open-source code and tokens. Its silver, round form resembles a pocket watch.
Initial setup involved plugging it into a PC, using Claude with Meta’s GitHub, and pairing the device with the Muse phone app. But the default experience was limited. Muse treated spoken commands as audio to transcribe, replies could be slow, and the pendant showed only a few words at a time without a full conversation view.
The basic software offered one Jolly avatar. It did not automatically find smart home devices, and a sample phone-camera feature lacked a remote shutter button. As a result, the device had little practical use beyond its initial setup.
Adding features through code
Claude Code helped fill those gaps. The owner added a conversation history page that opens with a swipe, along with an ElevenLabs British voice that reads Muse’s replies aloud. The voice was chosen from ElevenLabs’ free tier, though it did not match the cuddly look of the avatars.
Camera controls turned the pendant into a remote viewfinder. A button can trigger a picture on the phone, with a live preview; the phone’s front-facing camera image was also corrected so it no longer appeared upside down. A lower-quality preview over Bluetooth made it possible to check creative angles without relying on home Wi-Fi.
The project also expanded beyond assistance. Claude generated several avatars and a Tamagotchi-like game in which they need food, play, sleep, and a bath. The owner added four levels each for hunger, boredom, and dirt, plus consequences: treats leave a crunchy mess, while too much play makes a pet tired.
Small usability problems could be addressed in the same way. When releasing the push-to-talk button too soon sent nearly empty audio, the code was changed to send only recordings longer than one second. The remaining noticeable delays came from Meta’s servers, which were sometimes responsive and sometimes slow.
A pocket watch and a development experiment
Once the device was programmable, it became a platform for experiments beyond its original purpose. Claude built a pocket-watch face with three hands that tells the time. It also found, downloaded, trimmed, and loaded video clips of Zuck introducing the Muse Charm, then added a video player.
These additions illustrate the difference between a development kit and a finished consumer gadget. The Waveshare’s usefulness came from being reprogrammable; many of its most appealing functions were not part of the initial Muse setup. That flexibility may matter more to people who want to shape a device around their own routines than to buyers who expect a polished assistant out of the box.
Freedom comes with trade-offs
The owner says they have no coding experience and cannot judge whether the generated code is sound. That makes the project an impressive demonstration, but not proof that the device is dependable or safe for every use. The owner also did not trust Meta or Muse with email or access to a home network, and kept the gadget on guest Wi-Fi.
There is also a practical question: how much does a separate pendant add when a phone already offers Muse? The pendant can provide dedicated controls and a custom interface, but it is bulky and can be dropped. Without the extra code, it offered less functionality than the phone.
The broader test for the Muse Charm will be whether its own software and hardware let owners customize it this freely. A development kit gives curious users room to build, but most people may value a device that works well without programming. The experiment makes the case for programmable AI hardware while leaving open whether a consumer product can combine that freedom with reliability and clear privacy boundaries.