What the summary contains
A language model on your PC reads the transcript and writes:
- In short – the conversation in a few sentences
- Decisions – what was agreed
- Tasks – with owner and due date, when they were mentioned
- Open points – what is still unresolved
Every point carries one or more timestamps. Click one and you hear the passage it comes from. Templates for meeting, interview, lecture and podcast are included; in the settings you can write your own template that says what the summary should contain.
Ask your transcript
“Who sends the newsletter?” – Lautnote answers only from the transcript and shows the passage as a timestamp. If the answer is not in it, Lautnote says so instead of guessing.
How fast, how good
On our test PC (RTX 3060 Ti) the summary of a one-hour meeting took 28 seconds with the smaller model and 44 seconds with the larger one; a question takes a few seconds. Without a graphics card, the smaller model took about 23 minutes for an hour.
PCs with at least 8 GB of video memory get the larger model (Qwen3 8B, 4.7 GB), all others the smaller one (Qwen3 4B, 2.3 GB). In our tests the larger model found clearly more tasks with owner and due date. Small language models can make mistakes, so every summary says “may contain mistakes” – and the timestamps let you check every point in seconds.
The model is downloaded once, the first time you create a summary or ask a question, after you agree. Your recordings are not uploaded.