hermes-whatsapp-voice

hermes-whatsapp-voice is an open-source Hermes plugin that lets you call your Hermes bot over WhatsApp, or have Hermes call you. It pairs a Rust sidecar built on whatsapp-rust with Gemini Live for speech-to-speech conversation. It works with the personal-number setup where WhatsApp is linked as a device, so no WhatsApp Business or Cloud API account is needed.

What It Is

A voice layer for the Hermes agent. The sidecar, hermes-wa-voice, links to the bot's WhatsApp number as an additional linked device, answers calls only from a configured owner number, and places calls only to that number. Call audio is bridged to Gemini Live for interruptible speech-to-speech. During a call, the voice model can use an ask_hermes tool to hand real work to Hermes through its API server, and an end_call tool to hang up.

Calling in both directions

You can call the bot's WhatsApp number and talk to it. Hermes can also ring you for important items such as urgent alerts, priority reminders or cron jobs needing a decision, using the whatsapp_call_owner tool. The tool reports whether the call was answered, declined or unanswered, so Hermes can fall back to another channel. When a call ends, Hermes digests the transcript and sends a short summary in WhatsApp.

Setup path

Install with hermes plugins install ashwanthbalakrishnan5/hermes-whatsapp-voice --enable, or let an agent follow the bundled setup skill. Then set the owner number, pair the sidecar with the bot's number via QR code or pair code (from the dashboard tab or the hermes whatsapp-voice pair command), enable the Hermes API server, and run hermes whatsapp-voice serve or install it as a systemd or launchd service. Prebuilt binaries are fetched on first run and checked against pinned checksums; building from source is also supported. Requirements include Hermes 0.21.5 or newer on Linux or macOS, a separate WhatsApp number for the bot, and a Gemini API key.

Tradeoffs to know

The README states that WhatsApp calling uses an unofficial, reverse-engineered protocol, that updates can break it, and that automated use may violate WhatsApp's Terms of Service. It supports one call at a time, with no group or video calls. Gemini Live connections last about 10 minutes and are resumed transparently, though a resume can fail and end a long call.