Transcribed locally, audio never leaves the device

Request: „Record this meeting — local recording, parallel transcript, markers by keystroke.“

1. A start dialogue pulls meeting, agenda, links and participants from the calendar
2. Recording runs locally in short segments; a local speech-recognition model transcribes in parallel on the laptop's GPU
3. During the meeting I set markers by keystroke: to-do, decision, important, question, note
4. The transcript is cleaned — echoes, word loops, duplicates — and speaker changes are marked
5. After stopping, the analysis command is ready, or the analysis starts automatically

Whoever takes notes in a meeting thinks along less; in any organisation, long sessions end up as a few keywords, and cloud transcription rarely sits well with data-protection rules.

Obtain everyone's consent, start and stop the recording, approve deletion of the audio.

The first version lost text at segment boundaries and produced a name echo of around 3 percent of all words through term lists — the second cuts word-accurately and filters echoes.

Any leader with many meetings and strict data-protection rules. Privacy and speed are not opposites; recording happens only with everyone's consent.

Good minutes usually fail because nobody writes while thinking. Here the recording runs locally on the laptop, and a speech model on the same machine transcribes during the meeting. By keystroke I mark to-dos and decisions live. An eight-hour session day was transcribed in under two hours of compute — without a single second of audio leaving the device. The first version of the tool lost text at segment boundaries and produced a name echo; the second cuts word-accurately. Privacy and speed are not opposites; recording happens only with everyone's consent.