snegireff/meeting-scribe

meeting-scribe is a personal fork of an app called Simple Meeting Scribe, a Mac app that transcribes your video calls entirely on your own computer, with no cloud service involved. When it detects a Google Meet, Zoom, Teams, or Whereby call opening in your browser, it offers to start recording both your microphone and the other participants' audio as two separate tracks, then transcribes each one using an on-device speech recognition tool called WhisperKit, and identifies different remote speakers using a separate on-device tool. Transcripts are saved as plain text and structured files in a folder in your Documents, and an optional on-device AI can summarize the meeting and pull out action items without sending anything over the internet. This particular fork adds several features on top of the original app. It replaces Polish language support with Ukrainian, including a glossary system to fix jargon the speech recognition might get wrong. It adds one tap export of meeting summaries to a Telegram chat, an Obsidian notes vault, or a pre filled email draft, none of which happens unless you turn it on yourself. It remembers a person's voice once so future meetings automatically label them by name instead of a generic label, using an on-device voice matching system. It also adds a full text search tool over all your saved transcripts that AI coding assistants like Claude Code can call, optional live on-screen captions during a call, and pulling attendee names from your calendar to suggest speaker labels. The author is explicit that this is a personal tool they built for themselves and are sharing, not a polished product: there is no cloud, no telemetry, no automatic updates, no official support, and no promise that anything stays the same between versions. Running it requires an Apple Silicon Mac, since Intel is not supported, along with the very latest version of macOS, Xcode, and a tool called xcodegen, plus around 15 gigabytes of free space if you want every optional AI model cached locally. It is written in Swift and licensed under MIT, the same license as the original project it is forked from.