Hardware note 01
Meta Muse Gadgets on M5Stack Core2.
The hardware revision matters. So does the stage that was tested.
Source-checked 2026-10-05 No hardware testing
Check v1.0 and v1.1 separately.
M5Stack Core2 v1.0: listed upstream. Listed in the upstream SDK at the pinned commit; this is not a support or reliability guarantee.
M5Stack Core2 v1.1: not supported. The current Core2 port explicitly excludes v1.1 (AXP2101); this is not a claim that the hardware can never be ported.
Start with the power-management chip.
The documented port distinguishes Core2 v1.0 with AXP192 from Core2 v1.1 with AXP2101. Check your exact board documentation or markings. If the revision is unknown, keep compatibility unknown rather than choosing v1.0 by guesswork.
| Revision | Documented status | What to do next |
|---|---|---|
| M5Stack Core2 v1.0AXP192 PMU; ESP32-D0WDQ6-V3; chip revision ≥3.0 required by config | Listed upstream | Listed in the upstream SDK at the pinned commit; this is not a support or reliability guarantee. |
| M5Stack Core2 v1.1AXP2101 PMU; identify via the back sticker or PCB | Not supported | The current Core2 port explicitly excludes v1.1 (AXP2101); this is not a claim that the hardware can never be ported. |
Already paired is not the same as first pairing.
The contributor behind upstream PR #39 reported speech-to-text working with an existing pairing state. Fresh BLE pairing and battery operation were not tested in that report.
Documented or reported
- The Core2 port was merged upstream.
- The contributor reported an already-paired voice-to-text flow.
Still not established here
- A clean, first-time BLE pairing.
- Battery-powered operation.
- Any independent hardware test by Board Notes.
An open historical issue is not proof that all current builds fail. A merged port is not proof that every reported issue is resolved. Issue #9 ↗
“No response” needs a stage, not a blanket fix.
PR #70 changed chat queue allocation to prefer PSRAM, with internal-memory fallback. Its author reported a Core2 task-startup problem. That is a specific memory/startup path, not a universal explanation for silence.
A request that uploads and transcribes successfully but gets an empty assistant reply is a different observation. Issue #87 reports that later stage on a Waveshare 1.75C. We do not label it fixed by PR #70.
Separate voice input, text replies and spoken output →