← Bryl
§ Glossary

What is an AI medical scribe?

An AI medical scribe is software that records a clinical encounter, transcribes the audio, and drafts a structured clinical note from the transcript. The clinician reviews the draft, edits if needed, and signs it into the chart. The point is to remove the post-visit charting load — typing notes from memory at the end of the day — without sacrificing what gets documented.

What an AI medical scribe actually does

Every AI medical scribe performs the same three tasks, in order:

  1. Audio capture. The session is recorded — usually with the clinician's phone, sometimes with a tabletop microphone or a telehealth integration.
  2. Transcription. Speech-to-text converts audio into text, typically with speaker diarization so the clinician's and patient's turns are labeled.
  3. Note drafting. A language model reads the transcript and produces a structured clinical note — SOAP, DAP, BIRP, EMDR, Intake, Progress, or a custom template.

What varies between products is where those three steps run, who has access to the data along the way, and how the note ends up in the patient's chart.

Cloud-based vs on-device — the architectural choice

There are two architectural patterns for AI medical scribes today, and the difference is privacy-relevant.

Cloud-based AI medical scribes

Most AI medical scribes — Mentalyc, Upheal, Heidi, Blueprint, Eleos Health, Twofold, Abridge, Nuance DAX, Suki — run transcription and drafting on the vendor's servers. The recorded audio is uploaded; the transcript and note are generated in the cloud and returned to the clinician. Compliance is managed under a Business Associate Agreement (BAA) between the practice and the vendor.

The cloud model has trade-offs: server-grade speech models tend to be slightly more accurate, integration with EHRs through APIs is easier, and the vendor can ship improvements quickly. The cost is that protected health information leaves the device — and a BAA is a contractual mitigation for a risk the architecture chose to take, not a guarantee of confidentiality.

On-device (offline) AI medical scribes

On-device AI medical scribes run transcription and drafting locally — on the iPhone, iPad, or Mac — without uploading session content to a server. The audio, the transcript, and the draft note all stay on the device. There is no BAA because the vendor never receives protected health information. Bryl is the on-device option for psychotherapists.

On-device scribes became viable in the last two years because models like Apple's Foundation Models and NVIDIA's Parakeet TDT v3 now run faster than real-time on a recent iPhone. The trade is that an on-device scribe usually does not auto-push notes into an EHR through an API — that would require an authenticated handle on the EHR, which would re-introduce a server in the loop. The clinician copies and pastes the signed note instead.

Why it matters for therapists specifically

Psychotherapy sessions contain some of the most sensitive content in healthcare: trauma narratives, suicidal ideation, family dynamics, custody-relevant disclosures, EMDR processing blocks. The asymmetry of harm from a breach is severe. A cloud-scribe breach in primary care exposes vital signs and prescriptions; the same breach in a therapy practice exposes a patient's inner life.

That asymmetry is why architectural choice — not BAA strictness — is the question that matters most for solo therapists and small group practices. The full architectural comparison lives on the Bryl vs Mentalyc, Upheal, Heidi, Blueprint, Eleos, Twofold page, and the technical attestation is on the security page.

Related terms