Skip to content

Features

AutoMinutes combines local speech recognition and large language models into a single workflow for meeting documentation.

Core workflow

Live recording

  • Record directly from your Mac’s microphone
  • Real-time pipeline: audio → Whisper transcript → LLM minutes
  • Stop and resume via the top toolbar; progress is tied to each meeting session

Import audio

  • Upload existing recordings instead of live capture
  • Useful for interviews, calls, or archived meeting audio

Meeting mode

When Meeting mode is enabled, the language model formats output for typical meeting minutes (topics, decisions, action items). Turn it off for a looser summary style.

Speech-to-text (Whisper)

ModelNotes
Whisper large-v3-turboHigher accuracy, larger download
Whisper turbo (quantized)Smaller and faster, good for lighter hardware
  • Language: Auto-detect, English, or Chinese
  • Runs 100% on-device via Candle / local inference
  • No API keys or network calls required for transcription

Minutes generation (local LLMs)

ModelBest for
Qwen2.5 7BChinese and English meetings
Llama 3 8BBalanced performance and speed
Gemma 7BGeneral-purpose summaries
Yi 6BChinese and English

Tune generation with:

  • Temperature — randomness vs. consistency
  • Top P / Top K — nucleus and top-k sampling
  • Maximum length — cap tokens per generation round

Export & output

From the home screen and History:

ExportFormat
Speech audioOriginal recording file
Speech textPlain transcript
Minutes textStructured summary

Minutes can also be saved as styled HTML (with app branding) for sharing or archiving.

Supplement & refine

After the initial minutes are ready:

  • Type follow-up instructions in the supplement box
  • Import reference files: .txt, .md, .html, .docx
  • The model merges new context into the existing minutes

TIP

Wait until the current generation round finishes before sending another supplement.

History

  • Browse all meetings with search and date filters (all, last week, month, quarter)
  • Status indicators: audio only → transcript → minutes complete
  • Delete meetings (associated audio files are removed from disk)

Settings

SectionWhat you can do
ModelsDownload, delete, and inspect local model packages
AppearanceLight / dark / system theme
GeneralUI language and regional preferences
FeedbackContact the developer

Privacy by design

  • No user accounts or cloud sync
  • SQLite database and files stay in the app sandbox
  • Models downloaded once, used offline indefinitely

For the full legal text, see Privacy Policy.

Roadmap (planned)

Some capabilities listed in product design are still in progress:

  • Export to PDF and Word from the UI
  • Online STT / LLM API providers
  • Global recording status bar across views

The Mac App Store build focuses on reliable offline record → transcribe → summarize → export.