Local Meeting Notes

A local-first macOS meeting notes app with a quiet ChatGPT-style sidebar and Notion-like meeting note view.

What works in this MVP

• Create meeting notes from a left sidebar.
• Open to a home screen for starting a new recording.
• Starting a recording immediately creates a new session.
• Editable meeting titles.
• Default titles use local date/time, for example Thursday, May 21, 10:00 AM.
• Two main tabs: Summary and Transcript.
• Theme options: Light, Dark, Solarized Light, and System. Solarized Light is the default.
• English meeting transcription for the first release.
• Microphone recording through AVAudioEngine.
• System audio recording through ScreenCaptureKit when macOS permissions allow it.
• Local JSON storage under ~/Library/Application Support/LocalMeetingNotes.
• Auto-save for metadata, transcript edits, generated summaries, and audio file references.
• Local search across title, transcript, and summary.
• Markdown copy.
• App-managed local runtime hooks for whisper.cpp and llama.cpp.
• Generated app icon for packaged builds.

Run

swift run LocalMeetingNotes

The first recording may trigger macOS microphone and screen recording permission prompts. Screen recording permission is required by macOS for system audio capture through ScreenCaptureKit.

Local transcription setup

For source builds, the app can use a local whisper-cli binary from whisper.cpp:

brew install whisper-cpp
export LOCAL_MEETING_NOTES_WHISPER_MODEL=/path/to/ggml-base.en.bin
swift run LocalMeetingNotes

For distributable builds, put the runtime here before packaging:

Runtimes/
  whisper/
    bin/whisper-cli
    models/ggml-base.en.bin

If whisper-cli or the model is missing, recordings are still saved locally and the note is marked Setup needed. You can paste or edit transcript text manually in the Transcript tab.

Local summarization setup

By default, summaries use an app-managed local runtime if present:

Runtimes/
  llama/
    bin/llama-cli
    models/meeting-notes.gguf

If no local summary model is available, the app falls back to a deterministic no-install Markdown summary so the UI still works during development.

See docs/MODELS.md for hardware-aware model tiers.

Distribution

See docs/DISTRIBUTION.md.

Current limits

• Speaker diarization is not implemented yet.
• System audio capture depends on macOS Screen Recording permission and may need the app restarted after permission changes.