Trained on the standards you work by.
nursemate-4b is our fine-tuned nursing model — built on the MedGemma 4B foundation and trained on thousands of pages of UK healthcare guidance, including the NMC, NICE and more. Answers that speak the language of UK nursing, from the people who made Florence.
Why nursemate-4b
Built different. Trained on what matters.
Most AI assistants give you generic advice. nursemate-4b was fine-tuned on the documents UK nurses actually practise by.
Trained on UK guidance
Thousands of pages of NMC, NICE, RCN and regulatory documents — so the answers reflect how nursing is actually practised in the UK, not generic international advice.
MedGemma 4B foundation
Built on Google's MedGemma 4B — a medical-specialist base model — then fine-tuned further for UK nursing practice. Medical by design, not by accident.
Safety-aware by default
Fine-tuned with regulatory boundaries in mind — it knows where clinical judgement, referral and escalation belong, and says so instead of guessing.
Privacy-first
Designed to run locally on your own hardware — patient details and questions never need to leave your device or your care home's network.
Always available
No internet, no problem. A local model means Florence keeps working through outages — the 3am question gets answered at 3am.
Made for nurses, by a nurse
Born in the Nursemate family — the same team behind Florence, led by a nurse based in London. Fine-tuned on the questions nurses actually ask.
See it in action
A conversation with nursemate-4b
Click through the tabs — real answers, the way Florence will deliver them with the new model under the hood.
1. Call for help now — escalate to the registered nurse in charge and the duty doctor immediately.
2. Repeat the observations — confirm the score and check airway, breathing, circulation.
3. Start a track-and-trigger response — increased frequency of monitoring, typically every 30 minutes or continuous if clinically indicated.
4. Document everything — the score, the actions taken, and the time of escalation in the resident's record.
NEWS2 alone never replaces clinical judgement — but a ≥5-point aggregate (or any single parameter at 3) triggers the urgent response pathway. Document that you escalated, and to whom.
What to do:
1. Check the INR 3–5 days after starting the antibiotic — that's when the interaction typically peaks.
2. Flag it to the prescriber and the pharmacist — a dose review may be needed.
3. Watch for signs of bleeding: unusual bruising, blood in urine or stool, gum bleeding.
4. Document the interaction concern and the monitoring plan in the MAR notes.
This is a monitoring flag, not a reason to withhold either medicine — never change or stop a prescription without the prescriber's instruction.
What I can help with:
• The general principles of safe opioid administration (right resident, right drug, right dose, right route, right time — and documented pain assessment before and after)
• Typical ranges and conversion tables for reference, clearly marked as educational
• What to monitor after administration: sedation score, respiratory rate, and when to escalate
If you're unsure about a specific dose, stop and ask the prescriber or the pharmacist. When in doubt, escalate — never guess with controlled drugs.
Under the hood
The spec sheet
What nursemate-4b is made of, honestly and transparently.
| Attribute | nursemate-4b |
|---|---|
| Foundation model | Google MedGemma 4B (medical-specialist base) |
| Fine-tuning | Further trained on UK nursing & healthcare guidance |
| Training sources | NMC code & standards, NICE guidelines, RCN guidance and related UK healthcare documents |
| Parameters | 4 billion |
| Format | GGUF (llama.cpp) — runs on modest hardware |
| Hosting | Runs locally — no cloud dependency for inference |
| Context window | 8,192 tokens |
| Privacy | Questions never leave your device / network |
| Availability | Coming soon to Florence |
How we compare
nursemate-4b vs generic AI assistants
Not all AI is trained on the same books. Here's what changes when the model is built for UK nursing.
| nursemate-4b | Generic AI assistant | |
|---|---|---|
| Trained on UK nursing guidance (NMC, NICE) | ✓ | ✗ |
| Medical-specialist foundation (MedGemma) | ✓ | ✗ |
| Knows where clinical judgement belongs | ✓ | Sometimes |
| Can run fully offline / on-site | ✓ | ✗ |
| Answers stay on your network | ✓ | ✗ |
| Built by a nurse, for nurses | ✓ | ✗ |
| Free to use on Florence | ✓ | Usually paid |
Questions
Frequently asked
Is nursemate-4b a medical device?
No. nursemate-4b is a record-keeping and education aid, not a medical device. It doesn't diagnose, prescribe or replace clinical judgement — it supports the nurse who is always the decision-maker. We're explicit about these boundaries in the model's training and in the product.
When will it be available?
Very soon. We're putting the finishing touches on the model and the integration with Florence. Watch this space — the announcement banner on the Florence page will flip the moment it's live.
What is MedGemma 4B?
MedGemma is a family of open medical language models built by Google, designed specifically for healthcare tasks. The 4B version is a compact model that runs efficiently on modest hardware. nursemate-4b starts from that medical foundation and adds UK-specific fine-tuning.
Will it work offline?
That's the plan. Because the model runs locally, Florence should keep answering even when the internet drops — a real advantage for night shifts and rural settings. We'll confirm exact offline behaviour when we launch.
Is my data used for training?
No. Your conversations with Florence are not used to train the model. The fine-tuning was done on published UK healthcare guidance — never on user data.
Ready to meet Florence's new brain?
nursemate-4b is coming soon to Florence — free, local, and trained on the standards you work by.