On-device transcription
WhisperKit · Apple Neural EngineWhisper large-v3 runs on the Neural Engine of the Mac you already own. Mic and system audio are captured and transcribed locally — nothing is uploaded, streamed, or queued to a server.
Your meetings.
Your data. Hush.
Local-first meeting transcription and intelligence for macOS. Record, transcribe, analyze, and search — entirely on your Mac.
Private beta — Apple Silicon, macOS 14+
SPK 01
00:12Let's lock the launch date before we scope anything else.
SPK 02
00:19Agreed — I'll own the release notes and get them to you Friday.
SPK 01
00:34Good. And we keep the compliance review as a hard gate.
Decisions
Launch date locked; compliance review is a required gate.
Action items
SPK 02 — release notes by Friday.
Meeting intelligence that never touches the cloud.
Otter, Granola, and Fireflies send every word of your meetings to someone else's servers. For law, finance, healthcare, M&A, and anyone handling sensitive work, that's a dealbreaker — not a footnote.
Hush runs the entire stack — transcription, speaker separation, analysis, and search — on the Mac in front of you. Same workflow the cloud tools sell, without your conversations leaving the building.
The whole pipeline, on your machine.
01 — record → 05 — recallSpeaker diarization
SpeakerKit · pyannote CoreMLEvery voice is separated and labeled, across the room mic and remote participants alike. Map a speaker once and Hush keeps the attribution consistent through the transcript.
Meeting analysis
Gemma 4 · Metal GPUA local language model writes the summary, action items, and decisions — each grounded in cited lines from the actual transcript, so you can trace every claim back to who said it and when.
Ask your meetings
Hybrid search + RAGFull-text and semantic search across every meeting you've recorded, with a retrieval-augmented answer layer. Ask “what did we decide about pricing?” and get a cited answer from your own history.
Offline & air-gap ready
Zero network dependencyModels cache locally and the pipeline never phones home. Hush installs and runs with the network cable pulled — deployable on isolated and air-gapped machines where cloud tools simply aren't an option.
“We collect nothing” — and the code proves it.
No telemetry. No analytics. No crash reporting. No accounts. The only network requests Hush makes are optional model downloads and an optional update check — never your meetings.
NEVER LEAVESAudio, transcripts, summaries, search index
NEVER LEAVESSpeaker names, meeting titles, your notes
OPTIONALOne-time model download (skippable / offline bundle)
OPTIONALUpdate check (can be turned off for air-gapped use)
Runs on the Mac you already own.
Chip
Apple SiliconM1 or newer — Neural Engine + Metal
macOS
14 Sonoma+ScreenCaptureKit system audio
Memory
16 GB minLarger models auto-unlock at 24 & 36 GB
Network
Not requiredAfter models are on disk
Keep your meetings yours.
Hush is in private beta. Tell us about your team and how you'd use it — we'll be in touch about access.