Skip to content
HardwareLab
Robots & RoversIntermediate · Our suggestion

Nous-Girl: StackChan robot with Hermes Agent memory

The creator: @alexcovo_ethOriginal post

A talking StackChan desk robot with faces, DJ mode and an optional Hermes Agent brain

Full build guide

By @alexcovo_eth

View original on GitHub View original on X
Posted
GitHub stars
35 (GitHub)
Nous-Girl: StackChan robot with Hermes Agent memoryWatch on X
Video by @alexcovo_eth

What you need

Difficulty
Intermediate · Our suggestion
License
MIT (tribute face excluded) (source)

Parts

  • M5Stack StackChan K151 (CoreS3, ESP32-S3) (board)
  • USB-C cable, A computer for the voice server (only macOS on Apple Silicon tested so far)
AI model or service
xiaozhi-esp32-server, Parakeet (MLX) speech-to-text, Phonon text-to-speech, Hermes Agent (optional), Cloud model or Ollama

Steps

Difficulty
Intermediate
  1. Back up the robot, then flash the current firmware (robot, boy or girl face)
  2. Point it at the official xiaozhi cloud, or start your own xiaozhi-esp32-server in Docker for DJ mode, memes, email and Hermes Agent
  3. Join the robot’s Xiaozhi-XXXX Wi-Fi from your phone and pick your 2.4 GHz home Wi-Fi
  4. Hold the screen 1.5 s > Server, enter your server address and restart
  5. Optional: set up Hermes Agent as the brain
  6. Say “Hey Nous-Girl”

Full instructions: View original

Like this build? Get the weekly builds

Every Thursday: the 5 best gadget hacks you can copy. Free.

Built something? Submit a build

Overview

Nous-Girl is open-source firmware (MIT; tribute face excluded) by Alex Covo for the M5Stack StackChan K151 (CoreS3, ESP32-S3). It talks back with animated faces and lip sync, answers to “Hey Nous-Girl”, lights 12 RGB LEDs, dances, and has DJ, camera and Actor modes. It runs end to end on the creator’s own robot and Mac, but the public builds are “not yet tested on hardware”.

What it does

  • Animated faces with lip sync and moods; three CC0 starter faces (robot, boy, girl), or make your own.
  • Wake words “Hey Nous-Girl” and “Hey Jarvis”; tap the screen to stop it talking.
  • 12 RGB LEDs, preset and recorded dances, and a 3-pad head sensor.
  • Music, DJ mode and internet radio, with a live 13-band equalizer and beat-synced colors.
  • Speaks English, Spanish, French, German and Portuguese, each with its own voice.
  • Camera: “what do you see”, face ID, and following faces.
  • Email from its own AgentMail inbox, read back and confirmed before sending.

Parts and requirements

  • Computer for setup: A Windows, Linux or Mac computer to flash the robot and run the voice server; only macOS on Apple Silicon is tested so far

How it works

The robot runs custom xiaozhi-based firmware. On first boot you point it at a server: the official xiaozhi cloud works with no server of your own. Running your own xiaozhi-esp32-server in Docker on your computer adds DJ mode, memes, email and the Hermes Agent brain. There, Parakeet (MLX) turns speech into text, Phonon voice servers speak the replies, and the words go to Hermes Agent (long-term memory and tools) or straight to a model in Ollama.

LayerComponentRuns on
RobotNous-Girl firmware StackChan K151 (ESP32-S3)
Voice serverxiaozhi-esp32-server, or the official xiaozhi cloud for first boot Your computer (Docker)
Speech-to-textParakeet (MLX) Your Mac
Text-to-speechPhonon voice servers Your computer
Brain A (hardest)Hermes Agent + any model: long-term memory and tools Your computer; model in the cloud or Ollama
Brain B (medium)Hermes Agent + Ollama: long-term memory and tools Your computer
Brain C (easiest)Ollama only: no memory between conversations Your computer

Setup effort

The creator rates the brain setups from easiest (Ollama only) to hardest (Hermes Agent + any model).

Limits and caveats

  • Beta with one maintainer: it runs end to end on the creator’s own robot and Mac, but the public builds are “not yet tested on hardware”.
  • Firmware: v3.6.7-dj7 (prerelease); fallback dj5; avoid dj6 (it can reboot).
  • First boot works with the official xiaozhi cloud; DJ mode, memes, email and Hermes Agent need your own server.
  • Only macOS on Apple Silicon is tested so far for the voice server.
  • The multilingual Phonon voice is CC-BY-NC-4.0 and gated: accept its terms, non-commercial use only.
  • The Nous-Girl tribute face is fan art, not under the MIT license; public builds ship the CC0 starter faces.
  • The patched firmware source for each version is not public yet; flash the release files.

Hardware

  • M5Stack StackChan K151 (CoreS3, ESP32-S3)
  • USB-C cable
  • A computer for the voice server (only macOS on Apple Silicon tested so far)

More ESP32 buildsMore M5Stack builds

Credits and links

Sources

  1. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L44
  2. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L45
  3. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L46
  4. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L47
  5. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L48
  6. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L49
  7. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L52
  8. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L76
  9. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L69
  10. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L58-L70
  11. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L71
  12. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L72
  13. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L34
  14. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L35
  15. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L36
  16. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L56-L86
  17. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L5-L7
  18. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/VERSIONS.md?plain=1#L5-L8
  19. x.com/alexcovo_eth/status/2115689635158307269
  20. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L94
  21. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/README.md?plain=1#L120
  22. github.com/dax8it/nous-girl-stackchan/blob/fe8b02342a382598de1faa5c9bc93144549ea9e9/VERSIONS.md?plain=1#L177

This page is a short summary written by HardwareLab. Build steps, code and photos belong to the creator; follow “View original” for the full project.

Featured? Add this badge

Built this? Paste the badge into your README. It links back to this page.

Featured on HardwareLab

Markdown

[![Featured on HardwareLab](https://hardwarelab.dev/badge.svg)](https://hardwarelab.dev/projects/nous-girl-stackchan/?utm_source=badge&utm_medium=referral&utm_campaign=launch_2w)

HTML

<a href="https://hardwarelab.dev/projects/nous-girl-stackchan/?utm_source=badge&utm_medium=referral&utm_campaign=launch_2w"><img src="https://hardwarelab.dev/badge.svg" alt="Featured on HardwareLab" width="188" height="20"></a>

Related builds

All 28 in Robots & Rovers
  • Vision & Cameras

    RICOH GR Live View Shooting

    Hot-shoe remote on a StickS3.

    • M5Stack StickS3 (ESP32-S3, 8 MB flash, 8 MB PSRAM)
    • RICOH GR II, GR III/IIIx or GR IV
    • 3D-printed hot-shoe adapter (STL and STEP in the repo)
    • +1 more
    Beginner · Our suggestion45
  • Robots & Rovers

    OpenCat / Petoi

    OpenCatESP32 runs the OpenCat quadruped robotics framework on BiBoard, an ESP32-based development board for legged robots with up to 12 servos.

    • Petoi kits
    • ESP32 board
    • servos
    Intermediate · Our suggestion5.6K