Paste a visit transcript. Get a SOAP draft in seconds. Production version runs Claude or GPT under a HIPAA BAA — no training on PHI, 30-day retention.
Try the demoClinicGPT is an AI scribe for clinicians. The demo on this page runs entirely in your browser with a regex template — it never sends your text to a server. The production product uses Claude or GPT-class models under a Business Associate Agreement, never trains on your PHI, retains transcripts for 30 days for support, and exposes an audit log of every generation. The demo is illustrative; the production behavior is the contractual one.
Built around the way clinicians actually document
End-of-encounter draft from a transcript or recording. Editable, never auto-signed. The model emits a draft; the clinician reviews and signs.
Pull-down ICD-10 candidates inferred from the encounter content. Suggestions, not auto-coding. Clinician reviews before billing.
The SOAP note is a draft you edit in place. Every change is captured; the original draft is preserved for audit.
Every generation is logged with model, prompt hash, and output hash. Reproducible for compliance review.
Runs entirely in your browser. No text leaves this page.
The production behavior — not the demo — operates under these rules:
The demo on this page does not invoke any of this — it is a client-side regex sketch. It exists to show the SOAP-shaped output, not to demonstrate the privacy stance. The privacy stance applies to the production product.
By default the production product routes to Claude (Anthropic) for SOAP drafting because of latency and structured-output quality. Customers can elect a GPT-4-class model on Azure OpenAI instead, or — for the most restrictive deployments — a self-hosted open model on customer infrastructure. All three options operate under written BAAs.
No. The BAA explicitly prohibits use of customer PHI for model training, fine-tuning, or evaluation. Both Anthropic's and Azure OpenAI's commercial APIs honor this contractually as of 2026.
Through FHIR R4 for modern EHRs (Epic, Cerner/Oracle Health, Athena), and HL7 v2 for older systems. We do not require deep custom integration to start — clinicians can paste a transcript or upload an audio file and copy the signed note back manually. Native EHR integration is available on the Practice and Enterprise tiers.
Per-clinician pricing in three tiers: Solo (single clinician, paste-and-go), Practice (group of clinicians, EHR integration, audit dashboard), Enterprise (custom retention, customer-managed keys, dedicated support, deployment options). Pricing is on the order of a fraction of a typical scribe-service contract.
It suggests ICD-10 candidates and a CPT E/M level based on the documented MDM and time. It does not commit codes. The clinician (or coder) accepts, rejects, or overrides every suggestion. We treat coding as a human-in-the-loop decision.
Real risk and the honest answer: model outputs are reviewed by the clinician before signoff. The product is designed around the assumption that the model can be wrong, and the audit trail makes review traceable. We do not market it as autonomous documentation; we market it as draft documentation.
We run a 30-minute walkthrough with sample charts and your EHR. No commitment.
AI scribes have gone from demo magic to clinic-floor reality. Here's what the phrase covers in 2026 — and what it still doesn't.
Four prompt structures we use for clinical summarization, what each one is good at, and the failure modes we hit in production.
CMS, state boards, and major payers tightened expectations around AI-assisted documentation this year. Here's what your notes need to say in 2026 — and what they don't.