Tortus

TORTUS is the most regulated product in the ambient-scribe category, and that is the frame for everything else I am about to say. Built by a London company founded in 2022, it is an ambient voice medical device that listens to a consultation and writes the clinical note, referral letter and coding into the electronic health record for the clinician to approve. It is deployed inside the NHS — ambulance hubs, emergency departments, outpatient clinics and general practice — and it has recorded more than 2.5 million consultations. It is certified as a UKCA Class IIa medical device, holds ISO 13485, ISO 27001, Cyber Essentials Plus, DTAC and DSPT, and publishes the kind of safety evidence most competitors only gesture at. In my testing — which for this product means a product walkthrough, the G-Cloud documentation, and the published clinical evaluations, because you cannot sign up and try it — TORTUS is the AI scribe I would trust first in a UK clinical setting, with the trade-off that it is a procurement, not a download.

Where it positions itself

The pitch on tortus.ai/hospitals is blunt: "Don't buy a scribe. Build a partnership." That framing is honest about the sales model. TORTUS sells to NHS trusts directly (through G-Cloud 15, live from 7 September, or the RM6200 dynamic purchasing system) and to GP practices through a distribution partnership with X-on Health, which puts its Surgery Intellect platform into 3,500+ practices. Named deployments include London Ambulance Service (a one-year ambient pilot across the telephone clinical hub, visited by the Health Secretary), St George's University Hospitals (the largest trial of its kind in an ED, evaluated by GOSH DRIVE for NHS England), Royal Devon University Healthcare (the first trust to fully integrate TORTUS with Epic), and Great Ormond Street (a nine-site evaluation covering over 17,000 clinical encounters).

Its own model team calls the core agent OSLER, and the distinguishing technical claim is that OSLER combines voice, vision and click-based interaction to act inside the record — reading what is on screen, not just transcribing audio into a text box.

What it actually does

The core is the ambient note: multi-speaker capture of the consultation, a structured note in the trust's own template, a hallucination check against the transcript, and an approve-and-write action that pushes the signed note into the EHR. Around that core sit three modes: Meeting Mode for MDTs, huddles and board rounds; Dictation as a replacement for legacy dictation services; and template customisation that is per-specialty and clinician-editable. There is also a Data Insights layer that structures and audits every encounter.

Integration is where the product earns its keep. TORTUS integrates directly with Epic and Oracle Health (Cerner), and reaches EMIS, TPP SystmOne and Access through the TORTUS SDK and partners. Launch is context-aware from inside the patient's chart, so the note starts with the right patient and the signed note writes back to the record. It runs on the trust's existing laptop or workstation, with a recommended omnidirectional USB microphone for consulting rooms and ward rounds, and there is a native iOS app for bedside and community use. The data posture is explicit: no model trained on customer data, no patient audio retained at rest, UK and EEA hosting.

Pricing is more published than almost anything else in this category, at least for NHS buyers. The G-Cloud documents I read list a small-scale pilot at £44 per licence per month for up to 20 licences, a limited three-month pilot at £15,000 for up to 100 licences with training sessions and one on-site visit, an unlimited three-month pilot at £30,000, and Enterprise at £200,000 per year for up to 100,000 hours of processed audio with unlimited licences and £2 per hour beyond that. The current site describes two models — a fixed per-clinician annual licence banded by deployment scope, and a volume-based enterprise contract — both fixed in advance, with no per-minute meter.

My testing notes

Two concrete observations. First, the safety architecture is real work, not a badge. TORTUS documents a verification layer that checks each generated statement against the consultation transcript, and the published result — a 75% reduction in major hallucinations, appearing in npj Digital Medicine — is exactly the kind of evidence I ask for and almost never get in this category. Competitors ask you to take note quality on faith; TORTUS points at peer-reviewed data and at CREOLA, its clinical-AI validation platform involving over 100 clinicians. When I compared documentation approaches across UK scribes, this was the single largest differentiator — not the transcription quality, which everyone claims, but the willingness to measure and publish error rates.

Second, the EHR write-back model is designed the way clinicians actually work. In the walkthrough, the pattern is: launch from inside the patient's chart so context is already correct, capture the consultation, review a structured note in the trust's own template, approve, and the signed note lands in the record. Compare that with scribes that export a text block for manual copy-paste: the copy-paste step is where transcription errors become clinical errors. The multi-department evaluation numbers support the workflow claim — St George's reported documentation time halved in the ED and 47 minutes back per shift with a 13.4% per-shift increase in patient capacity, and Royal Devon projects 15,000 more appointments a year across Devon. These are evaluations by third parties (GOSH DRIVE for NHS England, NIHR HRC with the University of Exeter), which makes them more credible than the usual vendor case study, though still not measurements I made myself.

A third smaller observation: the G-Cloud pricing document exists, is specific, and names the overage rate. That sounds like a low bar, and it is, but most ambient-scribe competitors will not tell you a number at any stage. Being able to model £200,000 against 100,000 audio hours before a sales call is a meaningful procurement advantage.

Limits

Accessibility is the big one. There is no self-serve path for a small private clinic — everything routes through trust procurement or the X-on partnership, and the smallest self-serve-ish option I found is the £44 per licence per month pilot limited to 20 licences and one or two departments. A GP wanting to try it this week mostly cannot, unless their practice is already on X-on. Template customisation, while present, is more limited than some competitors, and the deployment model assumes the trust invests in onboarding sessions and template co-development.

Second, the geographic and system boundaries are hard. This is a UK product built around NHS governance — DCB0129 clinical safety case, DSPT, G-Cloud — and it makes most sense inside UK record systems. There is no private-practice PMS integration documented, and US Epic customers should not assume this maps to their environment.

Third, the published pricing has a dating problem. The G-Cloud document I read carried offers valid until 31 March 2025, while the current site points to G-Cloud 15 live from 7 September; independent comparisons flag exactly this conflict. The structural numbers (enterprise £200,000/year, £2/hour overage) have been stable across sources, but confirm current figures in your procurement — I would not present these as today's quotes.

Fourth, evaluation honesty: I could not run TORTUS in a consultation. Everything above about note quality rests on published evaluations, the vendor's documentation, and the regulatory record. For a Class IIa device that is arguably the correct evidence standard — but it is a different basis than a hands-on test, and readers should weigh it accordingly.

Verdict

For an NHS trust or ICS choosing an ambient scribe, TORTUS is the first conversation I would have. The combination of Class IIa certification, published hallucination data, direct Epic and Oracle Health integration, and pricing you can actually model is unmatched in the UK market I surveyed. Rated 4.3 out of 5 on the rubric — core functionality is deep and clinically grounded (the 40% pillar), value is strong versus the productivity gains reported in independent evaluations but assumes enterprise budgets (20%), usability is good on existing hardware but gated behind procurement and onboarding (20%), the ecosystem of frameworks, partners and EHR integrations is mature (10%), and reliability signals are the best in the category with published evidence and regulatory certification (10%). My recommendation: trusts should push for the three-month pilot and scrutinise template fit for their specialties; small clinics should wait for a self-serve option or route through X-on.

Sources

• https://tortus.ai
• https://tortus.ai/hospitals
• https://tortus.ai/tortus-and-x-on-health
• https://assets.applytosupply.digitalmarketplace.service.gov.uk/g-cloud-14/documents/722631/287098019864081-pricing-document-2025-03-13-0635.pdf
• https://aihealthindex.io/compare/system-c-careflow-ambient-ai-vs-tortus
• https://whitefieldhealth.com/resources/best-ai-medical-scribe-uk
• https://motics.ai/guides/tandem-health-alternatives-uk
• https://comparescribes.com/blog/accurx-scribe-vs-tortus
• https://doi.org/10.1038/s41746-025-01670-7
• https://tortus.ai/the-shell
• https://shell.tortus.ai/blog/hardening-the-shell