Local Voice Notes
From its catalog listing
Voice memos become text on the machine that recorded them: no API key, no account, and no audio leaving the host. Capture from PipeWire or ffmpeg, run faster-whisper locally, and the transcript arrives as a note. Four tools (transcribe_audio, voice_record_start, voice_record_stop, transcribe_last) plus a speech-to-text provider registered through ctx.register_transcription_provider, so core transcribes only admitted messages when stt.provider is set to hermes_local_voice. pre_llm_call attaches the pending transcript to the turn; the legacy pre_gateway_dispatch path is opt-in and off by default because it runs before sender authorization. Ships model management (list supported Whisper sizes with disk cost and cache status, download explicitly) and a live capture path with a silence gate so ambient room tone is never transcribed.
Good to know before installing
all processing is local. Network access happens only when the user explicitly runs the model download command, which fetches weights from Hugging Face into the Hermes model directory; no telemetry, no automatic updates, no reads outside the Hermes home and the recording paths the plugin creates. Disclosure \u2014 transcription runs locally with faster-whisper; when the user configures a loopback job server URL, audio files are POSTed to it and its reply is used as the transcript; the transcribe tool and the pre_llm_call hook read any local audio path named in the chat, and with cleanup enabled the tool deletes only recordings the plugin created; the opt-in auto_transcribe_inbound setting transcribes gateway voice messages before sender authorization; models are read from the Hermes model directory, or , and are downloaded from Hugging Face only when the user runs the model download command; no telemetry, no automatic updates. Disclosure — transcription runs locally with faster-whisper; when the user configures a loopback job server URL, audio files are POSTed to it and its reply is used as the transcript; the transcribe tool and the pre_llm_call hook read any local audio path named in the chat, and with cleanup enabled the tool deletes only recordings the plugin created; the opt-in auto_transcribe_inbound setting transcribes gateway voice messages before sender authorization; models are read from the Hermes model directory, `~/.cache/huggingface/hub` or `HF_HOME`, and are downloaded from Hugging Face only when the user runs the model download command; no telemetry, no automatic updates.
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 hermes-local-voice - Said no? Turn it on later:
hermes plugins enable hermes-local-voice - 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
- soilgrowerx (Community)
- Needs
- No API keys declared
- Adds tools
transcribe_audiovoice_record_startvoice_record_stoptranscribe_last- Adds hooks
pre_llm_call- Needs Hermes
- 0.21.5 or newer
- Version
- 0.1.1
- In the catalog
- 2026-11-12
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