Dialogue analysis and conversation protocol

An ordinary summary cannot be checked: you read a retelling and take it on trust. Here every item carries a verbatim quote, and our own code searches your text for that quote again before showing it. No quote — the item is not shown at all.

Text is enough; no audio is required. But if the transcript kept its timecodes, each item gets a moment attached, which makes it easy to go back to that part of the recording.

If the lines are labeled, we use the labels. If not, participants are inferred from the wording: that is accurate for a two-person conversation, can be wrong with several people in the room, and such a protocol is marked accordingly.

Yes. Text from any transcription service, an SRT or VTT subtitle file, a messenger export or plain typed text all work.

A language model writes the protocol; where it runs is stated in the note under the tool. Before sending we replace passport numbers, tax and insurance IDs, account and card numbers, phones and e-mails with placeholders. Names, company names and addresses stay in the text. The conversation itself is deleted after the analysis — only the protocol remains.

No. In the consultation scenario that is explicitly forbidden: the tool records what was said and gives no medical or psychological conclusions. The clinical judgment belongs to the practitioner.

They come back. The cost is held when the job starts and taken only when a finished analysis is handed to you; a failure at any stage, a closed tab or a recording the model cannot read all return the hold to your balance. That is why the number drops before the work is done and then either settles or returns — what you see during processing is reserved, not spent. If a recording fails twice in a row, the problem is almost always the audio rather than the service: check that speech is actually audible over the background and that the file is not a silent track exported by mistake.