Your computer is the robot's brain
Compatible ESP32-S3 firmware connects to your Desktop over a local network. The robot handles audio and device tools; Desktop provides conversation models, ASR, TTS, and the companion runtime.
/03 A COMPANION IN THE REAL WORLD
Bring your companion into the real world.
NomiFun Xiaozhi Yuntai
Give your desktop companion a physical presence. Connect Xiaozhi Yuntai so the same companion can hear you, speak with you, and respond through expressions and movement. NomiFun Desktop carries its identity, memory, and capabilities.
MEET YOUR COMPANIONSay something.
Say something.
Get a response.
Voice, expressions, and movement—in the same companion's world.
Expression and pan-tilt illustration · See the real device in the video belowNOMI
BUILT AROUND YOU
A familiar companion.
A real voice. Real movement.
Bind a device to an existing companion and bring its familiar personality, knowledge, and memory to your desk. Xiaozhi Yuntai provides voice, expressions, and movement; your computer handles reasoning and action orchestration.
Give emotion a way to show
An OLED display, spoken feedback, and pan-tilt movement take interaction beyond text. Available capabilities depend on the board and firmware configuration.
Put movement in the agent's toolbox
Device-side MCP tools expose head state and motion capabilities to your companion. Actions work alongside conversation, knowledge, and other tools in the same workflow.
Open hardware. Build it yourself.
The repository provides board adaptations, protocols, and firmware-flashing entry points. Follow the wiring and calibration requirements to give your companion a body of its own.
IN THE REAL WORLD
See how it responds.
A real device demonstration from the project. Your experience may differ with hardware, firmware, and model configuration.
Usage & release notes +
Compatible hardware, firmware flashing, network setup, and companion binding are required. Speech recognition, synthesis, and conversation use the models you choose. When cloud models are selected, the configured audio or text requests are sent to those providers. Calibrate the servos according to your board's requirements before enabling them.
Read the repository guide