Muse gadget firmware on three Waveshare ESP32-S3 boards

Difficulty

Beginner

AI-buildability

4/5

Nearly all the work is ESP-IDF firmware that AI tools can write and adapt, with off-the-shelf Waveshare boards that need no soldering, though ESP-IDF setup and board quirks still need hands-on debugging.

Time

A weekend

Cost

$25–50

Estimates based on the original write-up. How AI-buildability is scored. The maker built this with AI tools: more vibe-coded builds. Building with an AI assistant? Connect it to Afterware (free MCP server).

This fork of Meta's Muse Gadget SDK adds support for three Waveshare ESP32-S3 boards: the ESP32-S3-Touch-LCD-1.85C in a round speaker enclosure, the round ESP32-S3-Touch-AMOLED-1.43C, and the square ESP32-S3-Touch-AMOLED-1.8. All three run the full Muse UI with an animated avatar, push-to-talk on the BOOT button, touch settings and OTA updates. The boards are community ports and are marked experimental.

The fork adds spoken replies through ElevenLabs streaming text-to-speech, optional pushed messages, touch volume by dragging on the avatar screen, and battery percentage with a time-left estimate. It also includes a display freeze fix cherry-picked from upstream and watchdogs turned back on so a stuck task reboots the board.

The avatar gets a set of reactions: shaking the board makes it dizzy on boards with a QMI8658 accelerometer, it dozes with a snore before the screen sleeps, wakes when the screen returns, and giggles when its face is rubbed or tapped quickly. The repository notes that the code was written with AI coding tools, and the character art belongs to Meta.