docker·

⏱ 90s · full stack

Not what the robot runs

The Wireless CM4 has no Docker and no room for it (disk 89 %). Compose is for a laptop next to a Reachy Mini Lite, or a dev box pointed at the robot over the network. The robot itself runs the personas bare-metal — see Systemd.

The recommended deploy: three always-on personas as containers, sharing one SQLite brain, all connecting to the Reachy daemon over host networking. The image is light — no DDS, no CycloneDDS, no librealsense build (unlike neon), because Reachy control is pure HTTP/WS to the daemon.

build & up·

cp .env.example .env && $EDITOR .env
make build      # build the tiny image
make up         # docker compose up -d  → voice + telegram + thinker

the three containers·

container	role
`tiny-voice`	always-on bidirectional voice (local mic → speaker, head-wobble on speech)
`tiny-telegram`	Telegram DM listener → spawns a telegram-persona agent per message
`tiny-thinker`	30s heartbeat → photo + one expression + telegram status + journal

All three mount .memory/mem.db — the shared cross-persona brain — and reach the daemon over host networking (REACHY_HOST from .env).

operate·

make ps           # container status
make logs         # follow all logs
make logs-voice   # just the voice persona
make restart      # restart the whole stack
make down         # stop everything

live voice control·

make mute         # silence the voice agent (persists in mem.db)
make unmute
make voice-status

boot on power-on·

make install-compose-service   # user systemd unit → docker compose up -d at boot

See systemd for the bare-metal alternative.