Live Translate

From its catalog listing
Real-time speech-to-speech translation on your own computer: anyone speaks in any language and you hear them in yours, in their own voice, about 2-3 seconds behind. Hears ~100 languages (Qwen3-ASR + Whisper), translates between 201 (Qwen3.5-4B, a local server, a subscription, or the model Hermes already uses through ctx.llm), and speaks with the speaker's cloned voice in 23 languages and natural voices in 600+. One-way, two-way conversations and meetings/streams through a virtual audio device (Zoom, Meet, Teams, Discord, OBS, with OBS closed captions viewers can switch on); echo cancellation, background voices ignored, several speakers each with their own language and voice. Tools translate_start/stop/status/languages, /translate, hermes translate setup|start|file|serve|languages|devices, and an agent skill; optional desktop tray app (separate download).
Good to know before installing
while translating it listens to the microphone or audio input you choose and plays to the output you choose; it never starts from a chat app or a scheduled job unless you enable allow_remote_start. Voice cloning uses a 5-second sample of each speaker on this computer, except with ElevenLabs voices, which upload each speaker's sample to ElevenLabs to create a cloned voice in your account that is deleted when the speaker changes or translation stops; temporary local copies are deleted when the session stops. Each session's captions are saved to plugin-data/live-translate/sessions/*.jsonl. hermes translate setup builds a separate Python environment with uv (plus three voice-model environments on Windows/Linux) under plugin-data/live-translate, installs packages from PyPI, PyTorch's index (Windows/Linux) and llama.cpp's prebuilt-wheel index (abetlen.github.io), installs uv from PyPI into the plugin's data folder if it is missing, and downloads 6-9 GB of model weights pinned to exact commits from their official Hugging Face repos into the standard Hugging Face cache. While translating it runs the engine and voice-worker processes (stopped on Stop or when the session that started them ends) and a local server on 127.0.0.1:7865 only, protected by a token kept in an owner-only file under plugin-data/live-translate, a Host check and an Origin check. No network while translating with built-in or local models; a cloud model or voice you choose receives the session's text or audio with your key, read only through Hermes's secret store. No telemetry. Licences: Apache-2.0/MIT models by default; OmniVoice (best preset) is CC-BY-NC-4.0; on CPU-only machines Kokoro uses espeak-ng (GPL-3.0).
From its catalog listing.
How to install
- Install it. Hermes asks for any API keys it needs, then asks whether to turn it on.
hermes plugins install live-translate - Said no? Turn it on later:
hermes plugins enable live-translate - Restart Hermes so it loads the plugin.
Some plugins also need a sign-in or settings: see Full setup & settings. Memory plugins ask whether to become your memory provider instead. Steps follow the official plugins guide (opens in a new tab).
- Category
- Chat apps & voice
- Maintained by
- WestWaters (Community)
- Needs
- No API keys declared
- Adds tools
translate_starttranslate_stoptranslate_statustranslate_languages- Adds hooks
on_session_finalize- Works on
- macos, windows, linux
- Needs Hermes
- 0.22.0 or newer
- Version
- 0.1.0
- In the catalog
- 2026-11-12
Known issues (from its catalog listing)
- First setup downloads 6-9 GB of models and builds its own Python environments; run hermes translate setup once (or press Set up in the app).
- Real-time voice needs Apple Silicon or an NVIDIA GPU; on CPU-only machines captions come first and the voice follows with a delay.
- Under plugins.isolation: host the hermes translate terminal commands are not available (Hermes skips plugin CLI commands in the host); the tools and /translate still work.
Using the Hermes desktop app? See where plugins live

Full setup & settings ↗Source code ↗
More in chat apps & voice
New Hermes Agent updates every day on X. Follow @HermesWatcher