Yes, you can integrate the Greek god Hermes with Home Assistant, but not as a single official add-on. Instead, you combine a few community integrations—chiefly the Mercury speech-to-text engine, the Hassil sentence parser, and a custom assistant pipeline—to build a voice-controlled system that answers queries and triggers automations with a Hermes-themed persona. This article explains the practical setup, the tools involved, and the common pitfalls.
What “Hermes” Means in Home Assistant
In the Home Assistant ecosystem, “Hermes” is not an official product. It is a nickname borrowed from the Greek messenger god, applied to two different things:
- Hermes (Rhasspy’s old name): The Rhasspy voice assistant project was originally called Hermes. Some older tutorials still use that name, which causes confusion.
- Hermes as a persona: Users often name their voice assistant “Hermes” to give it a fast, messenger-like personality. This is purely cosmetic—you set the wake word, response phrasing, and TTS voice.
So when someone says “Hermes Home Assistant integration,” they usually mean either the Rhasspy-based voice pipeline or a custom-named assistant using Home Assistant’s built-in voice features.
How the Voice Pipeline Works
Home Assistant’s voice system (introduced in 2023) processes spoken commands in stages. For a Hermes-style setup, you need these components:
- Wake word detection: Detects “Hey Hermes” or a custom phrase (using openWakeWord or a local engine).
- Speech-to-text (STT): Converts audio to text. The community often uses Mercury, a local, offline STT engine that works well with Home Assistant’s Wyoming protocol.
- Intent parsing: The text is matched against sentences you define (e.g., “turn on the kitchen lights”). Hassil is the built-in parser that understands these templates.
- Action execution: The matched intent triggers a service call or automation.
- Text-to-speech (TTS): The assistant replies using a voice like Piper or a cloud TTS.
You configure all of this in Settings → Voice assistants, where you create a new assistant and assign each pipeline component.
Setting Up a Hermes-Named Assistant (Step by Step)
Here is a practical path to get a “Hermes” assistant running on a Raspberry Pi or a home server:
- Install the Wyoming integration (via HACS or the add-on store) to connect local STT/TTS services.
- Add the Mercury add-on for offline speech-to-text. It works well in English and several other languages, but it is resource-heavy—expect 1–2 GB RAM usage.
- Add Piper add-on for text-to-speech. Choose a voice that sounds “messenger-like” (e.g., a clear, neutral tone).
- Create a new voice assistant in the UI, name it “Hermes,” and select Mercury as STT, Piper as TTS, and the default conversation agent.
- Define custom sentences using the
conversationintegration. For example, a sentence like “Hermes, what’s the weather?” maps to theweather.get_forecastsservice. - Set a custom wake word if your hardware supports it. openWakeWord has a “hey_jarvis” model, but you can train your own with a few samples.
This setup is entirely local—no cloud services required—which appeals to privacy-focused users.
Common Problems and Fixes
Users run into a few recurring issues when building a Hermes-style assistant:
- Wake word not detecting: Often due to microphone gain. Adjust the audio input level in the Wyoming integration settings.
- Mercury uses too much CPU: On a Pi 4, it can be slow. Consider running it on a separate machine or using a lighter STT like faster-whisper with a small model.
- Intents not matching: Your sentence templates must be exact. Use the built-in “Test” button in the voice assistant settings to debug.
- No response after command: Check that the TTS service is running and that the output device (speaker) is correctly configured in the assistant’s settings.
- Multiple microphones: If you have several satellite devices, ensure each one points to the same assistant pipeline.
Hermes vs. Other Voice Assistants
How does this compare to alternatives?
- Home Assistant Cloud (Nabu Casa): The official paid service adds a cloud-based STT (Whisper) and TTS, which is more accurate than local options but requires a subscription and sends audio to the cloud.
- Rhasspy: The older project that started the Hermes name is more powerful for advanced users but has a steeper learning curve and is less integrated with Home Assistant’s modern pipeline.
- Amazon Alexa / Google Home: These are far more polished for general queries, but they do not support custom wake words like “Hermes” and they send data to the cloud. Home Assistant’s local pipeline is slower but fully private.
For most users, the built-in assistant with Mercury and Piper is the sweet spot—it’s free, local, and customizable enough to feel like “Hermes.”
Costs and Hardware Considerations
Running a Hermes-style assistant adds minimal cost if you already have a Home Assistant server. The software is open-source and free. Hardware needs:
- Microphone array: A USB mic or a dedicated voice satellite like the ESP32-based “Atom Echo” (around $10–15).
- Speaker: Any powered speaker connected to the server via 3.5mm jack or Bluetooth.
- Compute: Mercury STT works on a Pi 4 (2–4 GB RAM) but is noticeably faster on an x86 mini-PC. If you use cloud TTS, you avoid local CPU load.
There are no recurring fees unless you choose Nabu Casa for cloud processing.
Safety and Privacy
Because everything runs locally, a Hermes setup keeps your voice data on your own network. That is a major privacy advantage over commercial assistants. However, be aware of a few safety points:
- Voice control of locks, doors, or heaters should require confirmation phrases (e.g., “Are you sure?”) to prevent accidental commands.
- The wake word may trigger on similar sounds (like “Hermes” in a TV show), so place the microphone away from speakers.
- If you expose the assistant to the internet (via remote access), use strong authentication—voice commands can otherwise be sent remotely.
Frequently Asked Questions
Can I use “Hermes” as a wake word? Yes, but not out of the box. openWakeWord supports custom training, but it requires recording several samples of yourself saying “Hey Hermes.” Alternatively, you can skip a wake word and use a push-to-talk button on a satellite device.
Does Hermes work with French or Spanish? Mercury supports multiple languages, but accuracy is best in English. For other languages, you may need to use a cloud STT like Whisper via Nabu Casa for better results.
Is this the same as the Hermes protocol from Rhasspy? No. The old Rhasspy “Hermes” MQTT protocol is deprecated. Modern Home Assistant uses the Wyoming protocol instead. Don’t follow old tutorials that reference Hermes MQTT—they won’t work with current versions.
Can I make Hermes answer like the god? Yes, you can customize the conversation agent’s prompt (if using a local LLM) or simply set fixed responses in automations. For example, when you say “Who are you?” it can reply “I am Hermes, messenger of the gods—and your home.”
Do I need a dedicated PC for this? No, a Raspberry Pi 4 or 5 works, but expect a 1–2 second delay between speaking and the response. A cheap x86 mini-PC (like an Intel N100) will feel snappier.
Is there an official Home Assistant add-on called Hermes? No. The official add-ons are Mercury, Piper, and openWakeWord. The name “Hermes” is just a persona you create yourself.
Related
- Zeus Ydoria Resort Kreta: Facts & Review
- Apollo Duck Scotland Sailing Yachts: A Guide
- Zeus Role in Creation Myths: Greek Origins
How pages are sourced and corrected: Sources and editorial policy.