A macOS menu bar app that watches Zoom's AI Notetaker in real time, detects when a meeting ends, and automatically generates structured meeting notes via Claude, OpenAI, Gemini, or Ollama — saved directly to your configured output directory (Obsidian vault, Desktop folder, or anywhere you like).
No screen scraping. No network interception. Reads a local file Zoom already writes to your disk.
Zoom's AI Notetaker ("My Notes") writes a live transcript to a WebKit IndexedDB WAL file on your Mac as each word is spoken. This app watches that file:
~/Library/Application Support/zoom.us/data/
UnifyWebView_Cache/WebKit/UnSigned/Default/MyNotes/Origins/
<hash>/<hash>/IndexedDB/
1CB477F679D6.../IndexedDB.sqlite3-wal ← transcript (live during meeting)
DDEC8414E29A.../IndexedDB.sqlite3-wal ← blocks/title store
The WAL stores UTF-8 strings readable with strings(1) — no special permissions required beyond normal file read access.
Flow:
Zoom meeting starts
→ WAL file appears and grows
→ Menu bar icon switches to "In Meeting"
Meeting ends
→ WAL stops changing
→ After idle threshold: icon switches to "Generating"
→ Transcript is parsed, deduplicated, sent to your configured LLM
→ Structured notes saved to configured output directory
→ macOS notification fires
→ Icon returns to "Idle"
A native Swift/SwiftUI menu bar app spawns a headless Python engine as a child process. The Python engine watches the WAL and emits newline-delimited JSON events; the Swift app renders state and handles user-facing concerns.
ZoomNotesApp (Swift) zoom_engine.py (Python)
┌──────────────────┐ ┌─────────────────────┐
│ Menu bar │ stdout │ WAL poll loop │
│ Settings UI │◄────────│ Idle detection │
│ Notifications │ stdin │ LLM summarization │
│ Keychain │────────►│ Note writing │
└──────────────────┘ └─────────────────────┘
API keys live exclusively in the macOS Keychain (service zoom-notes-assistant). The Swift app reads them and injects them into the Python engine's environment at launch — no key files on disk.
Two files are written per meeting to the configured output directory:
Meeting Notes/ (default: ~/Desktop/Meeting Notes/)
Notes/
YYYY-MM-DD/
Meeting Title.md ← structured notes
Transcripts/
YYYY-MM-DD/
Meeting Title — transcript.md ← raw deduplicated transcript
Notes file frontmatter:
title: "Team Sync"
type: meeting
source: zoom-notes
date: 2026-04-21
created: 2026-04-21T19:01:36
attendees:
- "Alice Smith"
- "Bob Jones"
transcript: "[[Meetings/Transcripts/2026-04-21/Team Sync — transcript.md]]"
daily_note: "[[Daily/2026-04-21]]"Note body sections:
- Overview — 2-4 sentence purpose and outcome
- Attendees — speaker names from transcript
- Topics Discussed — sequenced, specific bullets per topic
- Key Decisions — explicit decisions made (or "No explicit decisions recorded")
- Action Items — Owner / Task / Due Date table
- Open Questions — deferred topics (omitted if none)
- Notes — additional context (omitted if none)
If you use Obsidian, point the Notes and Transcripts folders at your vault's meetings folder via Settings → Output. The frontmatter format (source: zoom-notes, wikilink-style transcript and daily_note fields) is compatible with Obsidian companion plugins for attendee resolution and daily note breadcrumbs.
A common setup: clone this repo into YourVault/Scripts/zoom-notes/ and configure the output dirs to YourVault/Meetings/Notes and YourVault/Meetings/Transcripts.
This app sends full meeting transcripts (including any sensitive personnel, legal, or commercial content) to whatever LLM provider you configure. By default that's Anthropic.
If you need local-only processing, configure Ollama as the provider in Settings → API / LLM. Ollama runs entirely on your machine and never sends transcripts to a third party.
- macOS 13 (Ventura) or later
- Zoom desktop app with My Notes (AI Companion / AI Notetaker) enabled
- API key for your chosen LLM provider (or Ollama installed locally)
No Python install required — Python 3.12 is bundled inside the app.
- Download the latest release (
Zoom Notes-<version>.dmg) - Open the DMG and drag Zoom Notes into your Applications folder
- Launch the app — a menu bar icon appears
- On first launch, an onboarding window walks you through setup:
- Click Open Settings to configure your API key
- Pick your LLM provider in the API / LLM tab
- Paste your API key — stored in macOS Keychain, never written to disk
- Adjust output paths in the Output tab if needed (default:
~/Desktop/Meeting Notes/)
- Start a Zoom meeting with My Notes enabled — the icon changes to "In Meeting" automatically
git clone https://github.com/nickybmon/Zoom-Notes.git
cd Zoom-Notes
# Fetch the bundled Python runtime (arm64 + x86_64 universal)
./scripts/fetch-python-runtime.sh
# Open in Xcode and run, or build a release DMG:
./scripts/release.shrelease.sh requires Xcode command-line tools, a Developer ID Application certificate, and a stored notarytool keychain profile named zoom-notes-notarytool.
| Provider | Where |
|---|---|
| Claude (Anthropic) | console.anthropic.com |
| OpenAI | platform.openai.com |
| Gemini (Google) | aistudio.google.com/apikey |
| Ollama (local) | ollama.com — no key required |
A spend limit is recommended for paid providers.
All settings are managed through the in-app Settings window. They persist to:
~/Library/Application Support/zoom-notes/settings.json— non-sensitive preferences- macOS Keychain (service
zoom-notes-assistant) — API keys
Tabs:
- API / LLM — provider, model, API key, connection test
- Output — notes/transcripts directories, subfolder pattern, filename pattern, custom frontmatter
- Prompt — system prompt customization
- Advanced — poll interval, idle threshold, WAL DB prefixes, base URL overrides
If you launch the engine outside the Swift app (e.g. for CLI debugging), the engine still reads these env vars as fallbacks if Keychain is empty:
| Variable | Purpose |
|---|---|
ANTHROPIC_API_KEY |
Claude key |
OPENAI_API_KEY |
OpenAI key |
GEMINI_API_KEY |
Gemini key |
ZOOM_NOTES_OUTPUT_DIR |
Override notes output directory |
ZOOM_NOTES_TRANSCRIPTS_DIR |
Override transcripts output directory |
ZOOM_NOTES_USER_NAME |
Your Zoom display name (exactly as it appears in meetings) — boosts the correct meeting when you're double-booked or in back-to-back meetings |
zoom_notes.py is also runnable directly for inspecting WAL state:
./venv/bin/python3 zoom_notes.py --list # Detect meetings in current WAL
./venv/bin/python3 zoom_notes.py --dump # Print full transcript to stdout
./venv/bin/python3 zoom_notes.py --watch # Live-follow transcript during a meeting
./venv/bin/python3 zoom_notes.py --notes # Generate and save notes now
./venv/bin/python3 zoom_notes.py --notes --dry-run # Preview without savingThese commands read the same settings.json and Keychain entries the menu bar app uses.
- Zoom updates may break this. The WAL path and DB structure are internal implementation details — no stability guarantee. If detection stops working, update the WAL prefixes in Settings → Advanced.
- After a meeting ends, Zoom checkpoints the WAL (deleting or shrinking it). The engine handles this by persisting accumulated transcript entries to
~/.cache/zoom-notes/on every tick — notes are not lost even if the WAL disappears before the idle threshold fires. - The
<hash>folder name appears stable per Zoom account but could change on re-login or app update. Runpython3 zoom_notes.py --listto verify the path. - Transcription accuracy depends on Zoom's server-side ASR, not local processing.
ZoomNotesApp/ # Swift/SwiftUI menu bar app (Xcode project)
zoom_notes.py # Core: WAL discovery, transcript parsing, LLM calls, note writing
zoom_engine.py # Headless WAL poller spawned by Swift; emits JSON events
zoom_config.py # Settings + Keychain helpers (Python side)
scripts/
release.sh # Full release pipeline: archive → sign → notarize → DMG
fetch-python-runtime.sh # Downloads universal Python 3.12 into python-runtime/
dmg-assets/ # DMG background image and appdmg config
zoom-transcript-extraction.md # Research: how the WAL was discovered and mapped
CLAUDE.md # Project overview for AI coding assistants