Talk it out, get a tidy note

Thoughts on the move are easy to dictate and impossible to read back: a transcript is a stream of speech with no punctuation, full of false starts and repetition. Two steps turn it into a note you actually return to.

What is inside

1. Voice message — the entry. The file is attached to the turn and the output port carries it along with the text: no separate “fetch the file” node is needed.
2. Transcription turns speech into text.
3. A tidy note — the model. It does not summarise; it cleans up: removes filler, adds punctuation, pulls tasks into a list.
4. The finished note — the exit.

The validation panel shows a message → file warning on the first edge. That is expected: the entry has no separate port for files, so there is only one way to wire this.

What to set up for yourself

• The second step’s prompt is where all the value is. It currently asks for tasks as a list; add your own: “pull out decisions”, “keep only the numbers”, “translate to English”.
• The transcription model is named explicitly, because an empty field on a media node means “no model selected” and the step fails. You may change it; you may not leave it empty.
• Language. Transcription detects it. If you mix two languages, name the main one in the node’s parameters and accuracy improves noticeably.

What to connect

A Telegram bot, and that is all. Voice messages are one button away in Telegram, which is why the channel is not incidental: the same graph on a website would make a person record a file first.