Skip to main content

Generating a note from a transcript

J
Written by Jason Fried

Once a session is transcribed, the AI is one click away from drafting your note.

Steps

  1. Open the patient and click into the session you want to write a note for (or open the session panel from a freshly stopped recording)

  2. Click Generate notes

  3. Watch the note stream into the editor

The first note for any session takes a bit longer — the AI loads itself into memory. Subsequent notes are faster while the model stays warm.

What's happening behind the scenes

For most transcripts, generation is a single pass: the AI reads the entire transcript plus your modality, density, and quote settings, and writes the note straight into the editor token by token.

For long sessions (~15,000+ words of transcript, roughly 90+ minutes of dense talk), the AI uses a two-pass pipeline: first it extracts clinically significant facts from the transcript in chunks, then it synthesizes a note from those facts. You'll see an extraction progress indicator before the note streams in. This adds a minute or two to the total time but produces a complete note instead of running out of context.

Generation time

Transcript length

Pro tier (16+ GB RAM)

30 min session

30–60 sec

50 min session

1–2 min

90 min session (two-pass)

3–5 min

Generation is CPU/GPU-bound. Closing other heavy apps speeds it up. The app stays responsive during generation — you can keep working.

What ships with the note

A SOAP note includes Subjective, Objective, Assessment, Plan, and a Session Recap. A DAP note includes Data, Assessment, Plan, and a Session Recap. Each section is a single paragraph except Recap, which is a bulleted list.

The note also pulls in:

  • Your selected modality (CBT, DBT, ACT, etc.) — affects the clinical voice

  • Your density (brief, standard, detailed) — affects how long each section is

  • Your quote level (none, some, many) — affects how often direct quotes appear

  • Per-patient case overview if available — gives the AI context about who this person is

  • Per-patient custom instructions if you've set any

  • The last finalized note for this patient — gives the AI continuity context

Everything happens locally. No transcript or session data is sent to any external service.

What if the note is wrong?

Edit it. The note is a draft — the editor is fully editable until you finalize. You can also regenerate as many times as you want with the same or different settings. Most users land on a workflow of one regeneration with adjusted settings, then a few minutes of editing.

Did this answer your question?