Complete for nowDerProtokollProfi
Meeting transcription and summarisation, entirely on your own device
Situation
There is no shortage of tools for meeting summaries, but they all upload the audio to someone else's server. That needs a consent solution, a data processing agreement and an approval. A tool that never leaves the device needs none of that.
Approach
Desktop application for macOS and Windows built on Tauri 2 with a Rust core. Capture from microphone and system audio, speech recognition locally through whisper.cpp and through Parakeet-MLX on the Mac, speaker separation, summarisation and task extraction through a locally running model. Storage in SQLite, encrypted with SQLCipher. The model backends run as separate subprocesses, so a crash in speech recognition does not take the application down with it.
Outcome
Seventeen completed stages, from capture through the compliance layer and conference room mode to signed installers and a measurement setup for runtimes. In daily use.