Transcriber* • Local-First Meeting Recorder & AI Notes

A meeting ends, the recording is saved somewhere, and the decisions inside it start to fade. Rewatching is slow. A generic summary is not enough when someone needs to find the exact discussion, recover a commitment, or turn it into a task.

Most meeting-note products solve that problem by making the recording and its history part of someone else’s service. I wanted a desktop-first alternative: keep the audio files, meeting library, transcript, and search index on the user’s machine. When transcription or AI assistance is needed, the user brings their own provider keys.

Transcriber* is built with Electron 39, electron-vite, React 19, and TypeScript. The Electron main process owns storage and provider integrations; the UI reaches it through a typed preload bridge, keeping desktop capabilities out of the renderer.

The recorder combines microphone input with available system audio, preserving a continuous clock through pause and resume. A floating control keeps the recording accessible while the main window is out of the way. When the recording stops, Deepgram Nova-3 transcribes it with speaker labels, punctuation, and formatting. Each transcript segment is editable, speaker names can be corrected across the meeting, and a line can take the audio player straight back to that moment.

Gemini works from the transcript—not the audio—to generate a structured summary, extract checkable action items, answer questions about the meeting, suggest a title and tags, and translate without replacing the original text. Action items live separately from the summary, so regenerating it never clears work that has already been completed.

The local library is backed by better-sqlite3 and SQLite FTS5, making every transcript searchable alongside meeting titles. Folders, tags, pinning, a recoverable trash, and exports to Markdown, plain text, SRT, VTT, or JSON make the record useful after the call is over.