Intelligenceyou can audit.
Software that speaks up about a patient has to be able to show why. So here is all of it: where every finding comes from, what happens when you overrule one, how AI is used, where the data lives, and what Terralis will never do. A medical director can read it in five minutes.
It shows its working
Facts are extracted from source documents, verified against the original text, and reconciled across sources by confidence. Local brands are mapped to their generics.
- OpenFDA drug labels
- EMA product information
- Ghana FDA product data and leaflets
- RxNorm, for drug names
- Ghana Standard Treatment Guidelines
- Licensed references as they are added
Overrides, on the record
A clinician can go ahead despite a finding. Sometimes that is exactly the right call. Terralis asks for a reason and keeps it, with every alert shown, in an append-only log.
That log is the facility's proof of diligence, and it is how we tune alerts over time: too many warnings and people stop reading them, so alert fatigue is treated as a safety risk in itself.
How AI is used
Every documented contraindication, interaction and dose limit is a structured, tested rule, checked the same way every time. AI has two narrow jobs: a classifier identifies and classifies medicines, and MedGemma, a medical language model we run ourselves, turns verified findings into plain explanations.
- AI never supplies a clinical fact
- AI never chooses a medicine
- Identifiers are removed before any AI processing
- Generated text is labelled and linked to its rule
- Most explanations are generated once and reviewed by a clinician
- If the AI tier is down, the rule text is shown instead
Where patient data lives
Patient data is kept in Ghana. Drug knowledge and the AI tier run centrally on Terralis infrastructure and never receive patient identifiers: only drug names and identifier-free context such as age, sex and kidney function cross over, and signed, versioned rule bundles flow back down.
Clinics, pharmacies, Lookup users
Shared
Terralis-hosted, Ghana data centre
Hospitals
Private cloud
Dedicated single-tenant instance, Ghana data centre
Large hospitals that insist
On your server
The hospital's own server; knowledge updates arrive as signed bundles
Today, during early access, our servers are hosted by Hetzner in the European Union and hold no patient data. Patient-data hosting in a Ghana data centre is being set up before the first hospital pilot, and the privacy policy will say so when it is live.
Who owns it
The facility. Terralis processes patient information on the organisation's behalf and never sells personal data, de-identified or otherwise. Revenue comes only from licences and integration fees.
- Export in a standard format, any time
- Full export within 30 days of ending a contract, then deletion with a certificate
- Source-code escrow available for hospital contracts
Built with clinicians
Clinical workflows should not be designed by engineers guessing. Doctors and pharmacists in Ghana shape what appears at a glance, how findings are worded, which questions are worth asking, and when an alert helps rather than becoming noise.
What Terralis is not
- Not a diagnostic tool
- Not a prescriber or treatment recommender
- Not a hospital information system or medical record
- Not a pharmacy point-of-sale or inventory system
- Not for patients to use on their own care
What it does
It compares what is known about a medicine with what is known about a patient, and presents findings for a licensed professional to verify. It reviews what the clinician proposes; it does not recommend what to prescribe. Off-label prescribing is legitimate, so the wording is always "confirm intent", never "wrong drug".
Regulation
Data Protection Act, 2012 (Act 843)
Health data is sensitive, and we treat it that way: encryption in transit and at rest, each organisation's data kept apart, role-based access, audit logs. Our registration with the Data Protection Commission is in progress.
Data Protection Bill 2025
Not yet law, but we build to it now: patient data in Ghana, a data protection impact assessment, a named data protection officer, and plain disclosure of where AI is used.
Medical device status
Terralis runs in check mode only: it reviews what the clinician proposes and never ranks or suggests treatments. We are asking the Ghana FDA whether check-mode decision support needs registration as software as a medical device, before any hospital pilot.
Clinician in the loop
Every finding shows its basis so a professional can review it independently, in line with how non-device decision support is defined in the US FDA's 2026 guidance. Terralis is clinician-facing only.