Skip to content

the family·

⏱ 45s

TINY is one of three robots built on the same shared brain, different body. If you know one, you know all three — the cross-persona nervous system (memory + agent_log + voice_bridge) is copied verbatim between them. Only the robot tool layer changes.

robot body control layer mirrors
NEON Unitree G1 humanoid DDS / FSM-gated arms + legs g1.py, tools/g1_*
SCOUT rover ROS / velocity agent.py, tools/
TINY Reachy Mini Reachy SDK → daemon :8000 tiny.py, tools/reachy_* 
flowchart TB
  BRAIN["🧠 shared brain<br/>memory · agent_log · voice_bridge · telegram · dispatch"]
  BRAIN --> NEON["🦿 NEON<br/>Unitree G1 · DDS · FSM"]
  BRAIN --> SCOUT["🛞 SCOUT<br/>rover · ROS · velocity"]
  BRAIN --> TINY["🤖 TINY<br/>Reachy Mini · SDK · daemon"]
  classDef b stroke:#7a6aa8,stroke-width:2px
  classDef r stroke:#2e8b8b,stroke-width:1.5px
  class BRAIN b
  class NEON,SCOUT,TINY r

what's identical·

  • 4 personas — shell · voice · telegram · thinker
  • shared SQLite brain — .memory/mem.db, unified reasoning log, voice bridge
  • deploy shape — docker-compose stack + systemd units, boots on power-on
  • single source of truth — one agent factory, one tool list

what's different·

Only the hardware layer:

  • NEON gates walking/arms behind an FSM + a single-writer arm mutex; has DDS, CycloneDDS, librealsense in a heavy image.
  • SCOUT issues ROS velocity commands to a mobile base.
  • TINY talks plain HTTP/WS to the Reachy daemon; no FSM, no danger class, a light image. Personality lives in head + antennas + emotion library.