Yes — Nosht uses AI. The model is Anthropic Claude Haiku 4.5 (claude-haiku-4-5-20251001), used in three optional, clinician-reviewed features: AI shorthand fill, an advisory note-completeness check, and a typed-shorthand scribe in the mobile app. There is no voice recording, the shorthand is designed to exclude patient identifiers, and the model is not trained on your notes.
Yes. Nosht uses artificial intelligence in specific, optional, clinician-controlled ways. The model is Anthropic's Claude Haiku 4.5 (claude-haiku-4-5-20251001), called per request through Anthropic's official API. It is not a Nosht-built model and it is not trained on your notes. Three features use it: (1) AI shorthand fill — turn your typed shorthand into structured template fields; (2) an advisory note-completeness check inside Bulletproof Mode; and (3) a typed-shorthand-to-note generator in the Nosht mobile app. Everything the AI produces is a suggestion you review and can edit before saving. Nosht does not record or transcribe voice, and the shorthand is designed to exclude patient identifiers.
When you type shorthand such as 'UR6 DO comp A2 lido 1 cart RD', Nosht resolves it in three layers. Layer 1 is a deterministic extractor (regex and lookup tables — no AI) that instantly recognises tooth numbers, surfaces, restorative material, local anaesthetic, carpule count and rubber dam. Layer 2 fills remaining fields from your own previous choices on that template (adaptive frequency defaults — not machine learning). Layer 3 sends only the still-unresolved clauses to Claude Haiku 4.5, constrained to the active template's fields, and streams the suggested values back so chips populate progressively. Every value the AI returns is re-validated server-side against the template — anything outside the allowed options is dropped. You then review and edit every field. AI only pre-fills empty fields and never overwrites what you entered.
We are deliberately precise about this, because accuracy matters in a healthcare product. Genuinely AI (Anthropic Claude Haiku 4.5): the shorthand-to-fields step, the advisory note-completeness audit, and the mobile typed-shorthand-to-note generator. Smart but NOT AI (deterministic rules, no learning model): the shorthand recogniser, your personalised 'Learned' defaults, the extra weight applied when you correct a suggestion, your custom abbreviation shortcuts, and the 30-day recency weighting that keeps suggestions current. We describe the personalisation as 'adaptive' because it counts your own habits — it is not machine learning, it trains no model, and it never leaves your account.
AI shorthand fill is optional. It is available on the Pro plan, only within your registered scope of practice, on four templates today (filling, simple extraction, crown, inlay/onlay), and it unlocks per template after you have saved five notes on that template so suggestions reflect your own style. There is no voice recording, no ambient listening and no speech-to-text anywhere in the Nosht web app. The mobile app's scribe works from typed shorthand — not audio. The note is never auto-finalised: the AI assists, you decide.
Nosht wraps the AI in deterministic, medico-legal safeguards. It never auto-asserts diagnostic findings: fields such as caries, BPE, periodontal status, mobility and radiograph findings are blocked from AI fill (the 'NAD guard'), so the AI cannot record something as normal or absent on your behalf. Consent and diagnosis fields are excluded from AI fill. The feature is fail-closed — an unset scope, the wrong plan or hitting the rate limit disables AI rather than guessing. The Bulletproof Mode AI audit is advisory only: it flags possible gaps for your review and never blocks or rewrites a note; only a high-confidence, high-severity gap affects the note's defensibility verdict, and if the AI is unavailable a clean note still passes.
Your shorthand is processed per request under a Zero Data Retention agreement with Anthropic: the input is used in-flight to fill fields, is not retained by Anthropic, and is never used to train any model. The shorthand is designed to exclude patient identifiers (names, dates of birth, NHS numbers), and Nosht's usage log stores counts only — not your note content. The AI is Anthropic's model accessed via API; it is not 'Nosht's own model' and it is not fine-tuned on your data. See our Data Processing Agreement and Security pages for the full detail.
Yes. Nosht uses optional, clinician-reviewed AI powered by Anthropic's Claude Haiku 4.5. Its main use is AI shorthand fill: you type shorthand and the AI fills the structured fields of your dental note, which you review and can edit before saving. Nosht also runs an advisory AI note-completeness check in Bulletproof Mode. There is no voice recording or transcription, the shorthand is designed to exclude patient identifiers, and the model is not trained on your notes.
Nosht uses Anthropic's Claude Haiku 4.5 (model claude-haiku-4-5-20251001), called per request through Anthropic's official API. It is not a Nosht-built or fine-tuned model, and it is not trained on your data.
No. There is no audio recording, no ambient listening and no speech-to-text anywhere in the Nosht web app. Nosht’s AI works from typed shorthand, not from voice. This is the main difference between Nosht and AI voice scribes such as Kiroku or Heidi.
No. The AI only pre-fills empty fields as suggestions. It never overwrites what you have entered, never auto-asserts diagnostic findings, and never finalises a note. You review and confirm every field before saving — the clinician stays in control and remains responsible for the record.
No. Your shorthand is processed in-flight under a Zero Data Retention agreement with Anthropic: it is not retained by Anthropic and is never used to train any model. Nosht’s usage log stores counts only, not your note content, and the shorthand is designed to exclude patient identifiers.
No, it is optional. AI shorthand fill is on the Pro plan, available within your registered scope of practice, on four templates today (filling, simple extraction, crown and inlay/onlay), and it unlocks per template after you have saved five notes on it. You can use Nosht’s structured templates entirely without AI.
AI scribes transcribe a voice recording of the appointment into a note. Nosht has no audio at all — you type shorthand and the AI fills structured, GDC/FGDP(UK)-aligned template fields that you review. The deterministic template core means the same selections always produce the same note, no patient audio is sent to the AI, and the shorthand is designed to exclude patient identifiers.
Genuinely AI (Claude Haiku 4.5): the shorthand-to-fields fill, the advisory note-completeness audit, and the mobile typed-shorthand-to-note generator. Deterministic and not AI: the instant shorthand recogniser, your personalised 'Learned' defaults, correction weighting, custom abbreviations and recency weighting. The personalisation is adaptive frequency counting — not machine learning, and it trains no model.
Powered by nosht.io — start a free trial.