Product DesignPrivacymacOS

Private Meeting Memory That Stays on the Mac

Hush

StatusCase study in progress
RoleProduct Strategy / UX / Product Engineering
ClientIndependent Product
Year2026

Hush is a local-first meeting intelligence product for macOS. It records, transcribes, identifies speakers, and makes past conversations searchable on the Mac. People can review the captured words, correct generated analysis, and choose what to export or share.

Visit HushRequest beta access
Private Meeting Memory That Stays on the Mac
Hush

Useful meeting memory usually asks for too much trust

Most meeting tools make transcription and summaries convenient by sending sensitive conversations to someone else's infrastructure. That tradeoff does not work for every team, especially when the discussion is confidential, regulated, or simply private.

Hush began with a different premise: the computer in the room should be able to capture and organize the meeting without turning the meeting itself into cloud data.

The current Hush workbench in light mode, using the built-in Project Atlas sample and a fictional host. No real meeting data is shown.
Hush

One local workflow from recording to recall

The product brings several jobs into one private workflow:

  • Capture microphone and system audio
  • Transcribe and separate speakers on the Mac
  • Produce summaries, decisions, and action items grounded in the transcript
  • Correct generated analysis while preserving the original
  • Search past meetings and return cited answers
  • Keep the core experience usable without a network connection

The interface has to make sophisticated local processing feel calm and understandable. Privacy is part of the product behavior, not a settings-page promise.

Action items stay connected to the captured words through speaker names, timestamps, and citations. All meeting content shown is synthetic.
Hush

Product decisions shaped by the trust model

Local processing affects onboarding, model downloads, performance feedback, storage, search, and failure recovery. The product must explain what is happening without making the user learn the underlying machine-learning stack.

That constraint shaped both the application and its beta path: clear system status, traceable outputs, controlled distribution, and claims that can be demonstrated rather than assumed.

Generated analysis is a starting point for review. People can edit summaries, actions, decisions, and entities. Hush keeps the generated original available while using the edited version for reading, exports, and configured integrations.

A saved edit becomes the effective summary, with the generated original still available. This edit was made through the real interface using the fictional sample meeting.
Current state

Hush has a working product, a private-beta path, and a standalone public site. These screens show the current light-mode interface through the product's own synthetic sample meeting. The core processing stays local; exports and configured integrations remain deliberate user choices. Test findings and beta evidence can be added as they become ready to share.