HumlaMeeting notes that never leave your Mac
Transcription and summaries run on-device with Whisper on Apple Silicon. No upload, no cloud, no account.
Free · Open source · macOS 13+ (Apple Silicon)
Watch it transcribe, right here in your browser.
Whisper runs on this device — the model downloads once from Hugging Face, then the clip is transcribed locally and the words light up in time with the audio. Open your Network tab: your audio is never sent anywhere.
First run downloads a ~75 MB model (cached afterwards). The full app uses a larger, more accurate model.
Most meeting tools ship your audio to their servers. Humla can run the whole pipeline — transcription, speaker labels, and the AI summary — on your own machine, so a confidential client call stays on your disk.
On-device Whisper
Transcribe with Whisper running on your Mac's GPU via Metal. Free after a one-time model download — nothing is uploaded.
Offline speaker labels
Diarization runs locally too, so you get colour-coded “who said what” without sending audio anywhere.
Your keys, or none at all
Want a cloud engine for speed? Bring your own OpenAI, Deepgram, or Groq key. Or go fully offline — your call, per language.
Local storage
Notes live in a local SQLite database on your disk. No vendor retention, no telemetry, nothing phoning home.
Works on a plane
The fully-local path needs no connection. Record, transcribe, and summarise entirely offline.
Open source
It's MIT-licensed and public, so you can verify the “nothing leaves your Mac” claim line by line.
See a meeting become a note.
New note, record, summary — that's the whole workflow. The cleanup afterwards is none, on a five-minute meeting or an hour-long one.
FAQ
Is meeting transcription really local?+
Yes. With on-device Whisper, audio never leaves your Mac — transcription and speaker labelling run on your GPU.
Do I need an internet connection?+
Not for the fully-local path. Cloud engines are optional if you want faster or higher-accuracy transcription.
Is it free?+
The app is free and open source. On-device transcription is free after a one-time model download; cloud engines use your own API key.