01
Realtime voice
Sub-second turn-taking. The robot listens through its mic; OpenAI Realtime handles ASR, reasoning, and TTS in a single WebSocket.
A voice-driven brain for Reachy Mini, backed by Nous Research's hermes-agent.
OpenAI Realtime drives the sub-second voice loop and robot motion — look, dance, antennas, camera. A single ask_hermes tool reaches a local hermes-agent for web search, calendar, and persistent memory. Cross-channel memory stays in sync via post-turn syncs.
hermes-body splits responsibility cleanly. The body listens and moves in real time. The brain remembers and reasons. A single tool call connects them.
Sub-second turn-taking. The robot listens through its mic; OpenAI Realtime handles ASR, reasoning, and TTS in a single WebSocket.
Eight robot tools — look, antennas, emotion, dance, camera, face_tracking, stop_moves, idle.
One ask_hermes tool reaches a hermes-agent container for web search, calendar, files, and long-term memory.
Four steps from a fresh robot to a talking, listening, moving body.
Run hermes-agent in Docker on a machine reachable from the robot. Enable its OpenAI-compatible API server and note the base URL and API key.
Setup guide →From the Reachy Mini dashboard — Apps tab, search "hermes-body" — or via the daemon API:
curl -X POST http://localhost:8000/api/apps/install \
-H "Content-Type: application/json" \
-d '{"url":"https://huggingface.co/spaces/wschenk/hermes-body"}'Set the OpenAI API key and hermes-agent endpoint via env vars on the robot. See the README for the full list.
Start the app from the dashboard. The robot listens on its mic and begins moving.