MeetingScribe
github.com
- Free plan
- Windows
- Web
What MeetingScribe is
MeetingScribe is a free, open-source desktop tool for capturing microphone and system audio during Google Meet or Zoom calls and creating transcripts, summaries, and action items. While a meeting is underway, a small faster-whisper model drafts a transcript locally on the CPU. Speaker identification uses key-free Resemblyzer by default; optional pyannote weights require a Hugging Face token. A local quantized language model can generate summaries and decisions, with an extractive fallback that needs no API keys. The project describes the app as local-first, keeping recordings, notes, and database files in a configurable local directory. MeetingScribe also offers editable notes and transcripts, templates, playback, highlights, comments, chapters, and talk-time analytics. SQLite FTS5 supports full-text search across past meetings, and users can ask questions across meetings with citations. Notes can be exported in several formats, and webhooks can post them to Slack, Discord, Notion, or Zapier. One limitation: live diarization labels speakers only as “Me” and “Others”; fuller diarization runs in the batch pass.
Who it suits
MeetingScribe suits people who want meeting notes for Google Meet or Zoom while keeping data on hardware they control. It may also suit users looking for searchable transcripts and multiple export options.
Good to know, and what to watch
- Captures microphone and system audio on desktop.
- Local models can draft transcripts and summaries.
- Includes cross-meeting search with citations.
- Exports notes in Markdown, text, subtitles, HTML, or JSON.
- MIT-licensed and free to run locally.
- Live diarization labels speakers only as “Me” and “Others.”
- Optional pyannote weights require a Hugging Face token.
Plans
Every plan and priceOpen source
Free plan
The verdict
MeetingScribe offers local-first capture, transcription, notes, and search for Google Meet and Zoom. Live speaker labels are limited to “Me” and “Others”; full diarization is done in the batch pass.
What the maker publishes
- AI notes
- A local quantized language model can generate summaries and decisions, with an extractive fallback that works without API keys. source
- Calendar integrations
- No source
- Cross-meeting search
- SQLite FTS5 provides full-text search across past meetings, and users can ask questions across meetings with citations. source
- Data handling
- The project describes itself as local-first and says recordings, notes, and database files stay under a configurable local data directory. source
- Exports
- Meeting notes can be exported as Markdown, plain text, SRT/VTT subtitles, a self-contained HTML page, or JSON. source
- Free plan
- Yes source
- Integrations
- A webhook can post notes to Slack, Discord, Notion, or Zapier when a meeting finishes. source
- Intended users
- The project describes itself as an open-source alternative for people who want meeting notes for Google Meet or Zoom on hardware they control. source
- Live transcription
- A small faster-whisper model drafts the transcript locally on the CPU while the meeting is in progress. source
- Meeting capture
- desktop source
- Meeting features
- It supports editable transcripts and notes, action items, meeting templates, audio playback, highlights, comments, chapters, and talk-time analytics. source
- Meeting platforms
- Google Meet,Zoom source
- Notable limitation
- The README says live diarization currently labels speakers only as “Me” and “Others”; full diarization runs in the batch pass. source
- Purpose
- MeetingScribe captures microphone and system audio during Google Meet or Zoom calls and creates transcripts, summaries, and action items. source
- Remote processing option
- Batch processing can optionally upload audio to a configured Cornell GPU cluster; if the cluster is unreachable, processing falls back to local CPU. source
- Scheduling
- Scheduled auto-recording supports Zoom OAuth and calendar feeds supplied as a secret ICS link. source
- Speaker identification
- Speaker diarization uses key-free Resemblyzer by default, with optional pyannote weights that require a Hugging Face token. source
- Summaries and action items
- Yes source
- Transcription
- Yes source
Checked against the maker's pages on 7 Oct 2026.