How Chronix Scribe uses AI, what it will and will not do, and where it runs. Chronix Scribe is provided by Chronix Health Group Ltd (company number 16663224, registered in England and Wales, registered office 11-13 Upper High Street, Epsom, Surrey, KT17 4QY, ICO registration ZB962513).
What the AI does
- Listens, with consent. Once the clinician confirms that everyone present has agreed, speech is turned into a written transcript by a speech-to-text service in Europe. Audio is never stored on our servers or by any provider; the optional offline mode of the Chronix Scribe mobile app (when the mobile app is released) holds audio encrypted on the clinician's own device only until it has been transcribed, and then deletes it.
- Drafts from what was said. A large language model drafts the clinical note in the clinician's chosen template, using only the transcript and anything the clinician typed.
- Links every clinical statement to its source. Each sentence that records a diagnosis, medicine, dose or plan is linked to the words in the transcript it came from, so the clinician can check it in one click.
- Helps with the paperwork around the note when the clinician asks: a patient summary in plain language (including other languages), letters and documents, suggested clinical codes, a short pre-visit recap from earlier notes the clinician chooses, and checks that the note's structure is complete (for example an empty required section) — never its clinical content.
- Translates in interpreter mode when the clinician turns it on. Every machine translation is labelled as machine-translated, and no audio is stored.
- Remembers writing preferences as short descriptors from a fixed list (for example "prefers concise notes") — never note content.
What the AI never does — attribute, never originate
- It records a diagnosis, medicine, dose or plan only when it was explicitly stated in the consultation.
- Anything it cannot link to the transcript is removed, and a visible marker shows where it was removed.
- It never originates a diagnosis, differential diagnosis, treatment option, risk score or calculation.
- It never fills the sections of a template that are reserved for the clinician.
- It does not make decisions about patients. It drafts documentation for a clinician to review.
- Chronix Scribe is not a medical device and does not give medical advice.
A clinician is always in charge
- Every AI-drafted note carries the label "AI-generated: review before signing".
- Every note stays a draft until a clinician reviews, edits and signs it. Nothing signs itself.
- Suggested clinical codes are labelled "AI-suggested codes — review and confirm" and are added only when the clinician confirms them.
- A patient summary is drafted from the note, and the clinician decides whether to share it.
Our support assistant is AI
When you use the support chat, or email support@chronixhealthgroup.com, you are first answered by an AI support assistant, not a person. It answers only from our help centre and refuses messages that contain patient details, which are not stored. It can make a refund when an organisation admin asks within the 14-day refund policy in clause 8 of the terms. You can ask for a person at any time, and a person reviews every email reply before it is sent.
Which models, and where they run
| Task | Service | Where |
|---|---|---|
| Speech-to-text | Speechmatics | Europe |
| Note drafting, patient summaries, translation, letters, coding suggestions, pre-visit recaps | Anthropic Claude models via Anthropic's API | United States (text only) |
| Support assistant | Anthropic Claude model via Anthropic's API | United States (text only; patient details refused) |
| Planned: note drafting inside the UK | Anthropic Claude models via Amazon Bedrock | London, UK — not in use yet |
The full list of providers is on the sub-processors page.
Your data and the AI
- Transcripts, notes and patient information are not used to train any AI model, ours or our providers'.
- Providers process the text only to return the result, under contracts that forbid other use.
Known limitations
- Speech-to-text can mishear words, names, medicines and doses, especially with background noise, strong overlap between speakers, or unfamiliar terms. Check medicine names and doses against the source link every time.
- The draft can attribute a statement to the wrong speaker.
- A removed-statement marker means the model wrote something it could not link to the transcript; decide yourself whether anything belongs there.
- Patient summaries in languages other than English should be checked by someone who reads that language before sharing where possible.
Tell us when it gets something wrong
Email support@chronixhealthgroup.com with the visit date and what went wrong (please do not include patient details). We review every report. To complain, see the complaints process.