Hermes Watch

What Hermes Watch does

You hold the right button, speak, and release. Audio streams as 16 kHz PCM over a WebSocket to a hub, goes through speech-to-text, the agent and text-to-speech, and the answer plays on the watch. A circular selector switches between five agent profiles, and a procedural avatar with a colour-coded status ring shows whether the agent is listening, thinking or speaking.

Other screens show quick commands such as STATUS, RECAP, STOP and NEW SESSION, a live reply ticker, and paged reply cards. Agent yes or no questions appear as approval cards answered with the watch buttons, and the answer is routed back to the agent. The Hermes Agent gateway is opened to ESP32-S3 gadgets through the hg.v1 wire protocol. Firmware updates arrive over WiFi with signed offers. The repository is licensed AGPL-3.0, and the README notes that the avatar engine is AGPL-3.0 code vendored from another project. An AGENTS.md file instructs AI assistants to read it before changing the project.

Key features

• Push-to-talk voice turns with 16 kHz audio
• Five agent profiles on a circular selector
• Procedural avatar with status ring
• Approval cards answered on the wrist
• Gateway plugin using the hg.v1 wire protocol
• WiFi OTA updates and a desktop simulator

When to use it

• Talking to a Hermes profile from a wrist-sized device
• Approving an agent question without opening a laptop
• Building an ESP32-S3 gadget that connects to the Hermes gateway

Who it is for: Hardware tinkerers and Hermes users who want a physical voice remote for their agents.

How it fits with Hermes Agent

It extends the Hermes Agent gateway to ESP32-S3 gadgets and is built around Hermes profiles.

Requirements: An M5Stack StopWatch (ESP32-S3 round AMOLED)

FAQ