fix: force short 1-2 sentence responses, reduce max_tokens to 80, cut playback wait

fix: drain mic buffer after each TTS playback to prevent feedback loop

fix: mute mic during TTS playback to prevent feedback loop (robot was hearing itself)

fix: decode MP3 via temp files, try ffmpeg/gstreamer/sox/mpg123 in order

fix: switch TTS from Kokoro (out of GPU quota) to edge-tts with Nigerian English voice

fix: add TTS logging to _speak() so we can see if Kokoro/playback fails

fix: switch VAD from mean energy to peak detection, threshold 0.08 — mean was too insensitive to speech

fix: raise silence threshold to 0.05 (background noise was 0.02), route listen loop logs to logfile

fix: route all logs to /tmp/storeos.log for debugging, add mic energy diagnostics

fix: use runtime_config.SAMPLE_RATE in transcribe()

Use Kokoro queue for TTS

Guard startup chain transactions

fix: add chain key, fix Kokoro TTS parsing, fix STT min duration

Add StoreOS runtime secret config

Fix Reachy app entry point

Enable Reachy local media by default

fix: remove custom_app_url to avoid port 7860 conflict with display server

write debug logs to /tmp/storeos.log + add /api/log endpoint

override wrapped_run to catch crash before run()

add custom_app_url, start display before main, better crash logging

add stderr prints to debug where app crashes

log stop_event state to debug early exit

wrap greeting in try/catch, add logging to prevent early exit

replace reachy_mini.utils.create_head_pose with inline impl to avoid scipy

pin numpy<2 and scipy<1.14 to fix circular import on Pi

remove 0g-storage-sdk dep — causes numpy/scipy conflict in shared venv

add crash logging to __main__

make chain/storage/inft optional so core app runs without heavy deps

hardcode API keys for hackathon demo

fix: add gcc + GObject/Cairo dev deps for PyGObject build

fix: Dockerfile use non-editable install

fix: colorTo must be from allowed list

StoreOS — decentralized AI store brain for Reachy Mini

initial commit