Building an AI Pocket Pendant for Fifty Dollars
In a column for theverge.com, Sean Hollister reports that rather than buying Meta’s upcoming Muse Charm—a squishy-faced AI companion designed to function like a Tamagotchi—he built his own version for $50 using a Waveshare ESP32-S3 1.75C touchscreen development kit, the Claude coding assistant, and Meta’s open-source developer tools.
The Waveshare hardware arrived as a silver metal pocket watch pendant drop-shipped from China. While it lacks built-in cameras or a 5G cellular radio, it connects directly to Meta’s Muse AI cloud using open-source code and authentication tokens.
Initial setup required plugging the device into a PC, feeding Claude the repository instructions from Meta’s GitHub, and pairing the pendant with the Muse phone app. Out of the box, however, the device suffered from functionality gaps. Meta’s software interpreted voice inputs as audio transcription clips rather than direct commands, responses were slow, and text output was restricted to a few words on screen without a scrollable conversation history.
Reprogramming the Waveshare Devkit With Claude Code
Hollister writes that a few sessions with Claude Code resolved nearly all of the hardware and software limitations. To fix the unreadable text replies and lack of audio feedback, Claude built a conversation history page accessible via a swipe gesture and integrated an ElevenLabs British voice model to read responses aloud.
Meta unveils Muse Charm, a tiny keychain sized AI device with a Tamagotchi-style displayThe new gadget is designed to give users a playful way to interact with Meta’s Muse AI, with Mark Zuckerberg describing it as a “joyful little device” pic.twitter.com/Y1ePyn1luG
The coding assistant also added functional camera shutter controls, allowing the pendant to trigger high-quality photos from across a room with a live phone preview over Bluetooth. Furthermore, Claude patched a bug that caused the phone’s front-facing camera display to render upside down.
To mimic the Tamagotchi-style interaction, Hollister asked Claude to generate a batch of custom avatars alongside a dedicated minigame. The digital pet features four distinct tiers of hunger, boredom, and cleanliness. Feeding the avatar treats introduces a crunchy mess mechanic that accelerates dirt accumulation, while excessive play drains the pet’s energy meter.
Timekeeping and Additional Customizations
Additional programming tasks completed in minutes included a fully functional pocket watch interface featuring three active clock hands. Claude also located, downloaded, trimmed, and installed video clips of Mark Zuckerberg introducing the Muse Charm, embedding a local video player directly onto the Waveshare hardware.
To mitigate accidental audio truncation caused by prematurely releasing the push-to-talk button, a quick code patch was implemented to filter out audio clips shorter than one second. Hollister notes that any remaining latency stems directly from Meta’s server infrastructure rather than local hardware processing.
Security Precautions and Hardware Independence
Reflecting on the experience, Hollister notes that he lacks formal coding background and cannot verify the safety or structural integrity of the generated code. Citing privacy concerns—particularly following reports of a reviewer’s home address leaking via Meta’s Muse agent—the device was isolated on a guest Wi-Fi network from the beginning.
Hollister concludes that the primary value of dedicated wearable hardware lies in custom programmability rather than replicating standard smartphone capabilities. Whether Meta’s official Muse Charm will offer similar open development pathways remains unconfirmed ahead of its retail release.
