meeting-transcriber: Transcribe macOS Meetings Locally with WhisperKit

Overview

Meeting Transcriber is a local-first meeting recorder and transcriber for macOS, built on WhisperKit from Argmax and NVIDIA’s Parakeet, with FluidAudio for speaker diarization. It captures meeting-app audio and microphone audio separately, processes recordings on-device, and can produce structured Markdown notes using Claude Code CLI or an OpenAI-compatible endpoint.

It is aimed at people who want meeting transcripts without sending audio to a cloud transcription service, especially in privacy-sensitive or offline settings. It supports Teams, Zoom and Webex detection, with additional opt-in detection paths and file imports for existing recordings.

Key Features

• Automatically detects Teams, Zoom and Webex meetings and records app audio alongside microphone input.
• Transcribes locally with WhisperKit, which supports 99+ languages, or NVIDIA Parakeet TDT v3, which supports 25 European languages.
• Separates speakers with FluidAudio, including separate diarization of app and microphone tracks.
• Generates meeting summaries, decisions and tasks in Markdown through Claude Code CLI or an OpenAI-compatible service, or saves transcripts without notes.
• Imports common audio and video formats, with ffmpeg required for MKV and WebM.
• Offers record-only mode and a localhost automation API in the Homebrew build for external processing and integrations.

Use Cases

• Teams handling confidential, legal or healthcare discussions can keep recordings and transcription on a Mac rather than uploading audio to a meeting service.
• Individuals who need searchable meeting records can capture supported calls and generate notes with a configured local LLM or Claude Code CLI.
• Researchers or multilingual teams can select WhisperKit or Parakeet based on language needs and process imported recordings.
• Organizations with their own processing infrastructure can use record-only mode to capture source tracks for a separate pipeline.

Getting Started

Install with Homebrew:

brew tap pasrom/meeting-transcriber
brew install --cask meeting-transcriber

The app requires macOS 14.2 or later. See the README for permissions, provider setup and build instructions.

Alternatives

• meetily: Meetily records and transcribes meetings locally and can use Ollama or other providers for AI summaries, unlike this app's WhisperKit or Parakeet workflow.
• mityu: Mityu emphasizes source-linked summaries and action items, while this app focuses on macOS call recording, on-device transcription, and speaker separation.

Considerations

• The project is young, despite frequent commits and releases. Recent issue closures and merged pull requests indicate active maintenance, but its very high contributor concentration makes continuity a consideration for teams adopting it long term.
• The app requires macOS 14.2 or later for its app-audio capture approach. Microphone access is needed for mic recording, and some detection or participant features rely on additional macOS permissions.
• Transcription models download on first use. The README describes WhisperKit’s stock model as roughly 1 GB and Parakeet’s as roughly 50 MB; initial setup therefore requires network access and local storage.
• Meeting-note generation depends on Claude Code CLI or an OpenAI-compatible API endpoint. The transcript-only mode avoids that dependency, but does not generate protocols.
• The README describes unit tests, linting and analyzer checks in CI, but the repository health data does not report tests as a measured requirement or capability.