Give your Hermes Agent a body.

This turns a cheap AliExpress AI robot toy into a voice device for Hermes Agent — running entirely on your own machine, using your own model, with no vendor cloud in the loop.

Hold one button, talk, and Hermes answers out loud. The robot's screen becomes Hermes's face, and the whole thing runs on its own battery.

That's it. No soldering, no extra parts, no subscriptions.

Three things happen: flash the robot, tell it your Wi-Fi, connect it to Hermes.

Install esptool:

pip install esptool

Download the firmware from firmware/ in this repo, then flash it:

esptool.py --chip esp32s3 --baud 460800 write_flash 0x0 hermes-gadget-spotpear-128-box-0.2.0.bin

Plug the robot into your computer with the USB-C cable when you run that. If it won't connect, unplug it, hold the boot button (the one to the right of the USB-C port), plug it back in while still holding, then release.

Full walkthrough →

The robot has no keyboard, so it borrows your phone:

1. Plug the robot in and wait for it to show a setup screen.
2. On your phone, join the Wi-Fi network called Hermes-XXXX — the password is shown on the robot's screen.
3. Open http://192.168.4.1 and enter your home Wi-Fi details.

Your password goes straight from your phone to the robot. It never passes through any server.

On the machine running Hermes:

hermes plugins install https://github.com/Adolanium/hermes-gadget-sdk/tree/main/plugin --enable
hermes config set platforms.gadget.enabled true
hermes gateway restart
hermes gadget info

Then allow the gadget port through your firewall — this is the step most people miss:

# Linux (ufw)
sudo ufw allow from 192.168.1.0/24 to any port 8765 proto tcp

The robot connects within a few seconds and shows Ready.

Hold the boot button, talk, release. Hermes answers out loud.

Full walkthrough →

Speech recognition and text-to-speech use whatever you already have configured in Hermes.

• Port 8765 reachable from your LAN. Bluetooth isn't used.
• Speech configured. Speech-to-text and text-to-speech, either through hermes setup or the stt: / tts: sections of config.yaml. Without it the robot still works, but replies are text-only.
• ffmpeg if your speech output isn't WAV.

The board profile for this exact robot is in firmware/spotpear-128-box.patch — a patch against the Hermes Gadget SDK.

For a different robot, skills/hermes-gadget-hardware/ is an agent skill that walks through identifying unknown hardware and porting it — reading the vendor's schematic and config, choosing the right driver enums, and knowing which parts need new code versus configuration. Point your agent at it.

This builds on other people's work:

• Hermes Gadget SDK by Adolanium and contributors (MIT) — the firmware and the Hermes plugin. Independent community project, not affiliated with Nous Research.
• xiaozhi-esp32 — the open-source firmware the robot ships with. Its board definitions and SpotPear's published schematics are what made the port possible.
• SpotPear — the robot itself, with schematics and pin tables published.
• Hermes Agent by Nous Research.

The firmware in firmware/ is MIT, derived from the Hermes Gadget SDK. See LICENSE.

Built by Hermes.
_{Researched the hardware from a product listing, identified the board over the network, read the vendor's own schematic and firmware source, wrote the board port, fixed three bugs in the SDK to make it build, compiled and flashed the firmware, and wired the device into the gateway — start to finish, in one session.}