An orchestration agent, scoped with you before it is priced. This agent coordinates several specialised agents. Its orchestration follows your actual workflows — which is why it is scoped with you rather than bought off the shelf. We establish the scope together, then the quotation commits it.
The journeys described below form the scope that this review refines and the quotation commits. The specialised agents it coordinates can be ordered today.
Request a quote
● Public sector — Care & health
The AI agent for nurses: paperwork in the service of care, never the other way round
Handover notes, keeping the care record and planning rounds take up a considerable share of nursing time — time taken away from the patient. Your AI agent writes up, structures and keeps things current, from your notes or from dictation. Hosted in France — on local inference or an isolated resource, the HDS requirement settled at scoping — health data stays under control. The nurse keeps the clinical decision.
Hosting in France
Health data protected
GDPR & AI Act: governed deployment
The carer decides
Updated on
Deployed in a few weeks
Nursing assistant · hosted in France
Mrs Lefèvre, room 214: dressing changed at 2pm, wound clean, pain rated 3, vital signs stable. Write that up properly in the handover notes.
Handover note written in the focused format (focus / data / action / response): dressing changed at 2pm, wound clean with no sign of infection, pain 3/10, vital signs within range.
I have carried over the vital signs from the last set of readings and flagged that the next dressing should be scheduled for day 2.
⛓ Source · the patient's care record + the vital-signs readings
I have carried over the vital signs from the last set of readings and flagged that the next dressing should be scheduled for day 2.
⛓ Source · the patient's care record + the vital-signs readings
And for my round tomorrow morning, where do I start?
Suggested order for the round covering 7 patients, optimised around the care with fixed times (two injections at set hours, one complex dressing to allow time for).
To be adapted to your clinical assessment on the day — I prepare the list, you decide.
✎ Action · round plan proposed — the nurse approves
To be adapted to your clinical assessment on the day — I prepare the list, you decide.
✎ Action · round plan proposed — the nurse approves
Local inference · no data outside the EU
Health data hosted in France
Sovereign by designLocal inference or hosting in France
GDPR & AI Act: governed deploymentTraceability & human oversight
TurnkeyDesigned, installed and operated for you
The nurse decidesThe agent assists, it never delivers care alone
✦ In brief
For a care ward or a nursing team, a Blue Lemon Agent agent absorbs the documentary load — writing up handover notes, keeping the care record, help with planning rounds — from notes or from dictation. It runs on local inference or is hosted in France in a dedicated, isolated environment: health data is never exposed to a foreign service, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. The time gained is given back to care and to patients. The agent assists, the nurse decides. Live within a few weeks.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
9
uses ready to deploy on this scope
0
decision taken without human approval
Reference points describing our offer, not results measured at a client. The scale of the gain is confirmed by a pilot on your own scope.
The context
Why AI matters to care wards — and why they hesitate
Patients and the institution expect quality care and unbroken continuity. But the time available is mechanically reduced by handover notes, keeping the record and coordination — and the data involved is among the most sensitive there is.
! The issue
A ward is caught between a requirement for quality and traceability of care and a documentary load that keeps growing (handover notes, care record, planning). Yet most consumer AI tools amount to entrusting named health data to a third party, often hosted outside Europe and subject to the Cloud Act — incompatible with medical confidentiality and with the HDS requirement that applies in France to hosting health data.
✓ Our answer
In care, AI is only of interest if it is sovereign and confidential by design. Local inference or an isolated resource hosted in France, the HDS requirement settled with you at scoping, systematic human oversight, clinical decisions reserved for the professional: time gained on documentation is never paid for in lost confidentiality. The aim is not to replace the nurse, but to give them back time for the patient.
The decisive point
The confidentiality of health data: sovereignty & compliance
A care ward handles the most sensitive data there is. Here is how the architecture of our agents protects it, record by record.
Local inference
The agent can run on a machine at the establishment: no patient data leaves the network, nothing passes through a public cloud.
Hosting in France, HDS requirement at scoping
Otherwise, a dedicated and isolated resource, hosted in France, under French law — your data: processing and access within the European Union targeted by the architecture. In France, hosting health data on behalf of others is subject to HDS certification of the host: that requirement is settled with you at scoping, before any go-live.
Reduced extraterritorial exposure
Exposure of health data to the Cloud Act and FISA 702 is reduced by design; location alone does not guarantee immunity.
One isolated resource per department
No pooling of health data: an environment strictly dedicated to your establishment.
Encryption & controlled access
Encryption in transit and at rest, role-based access, strong authentication and logging of access to the record.
AI Act: governed deployment
The agent is strictly in support; no automated clinical decision; traceability and human oversight from end to end.
What depends on the architecture chosen These points are not general guarantees: they are settled deployment by deployment, in the quotation.
- The applicable location is that of the architecture set out in the quotation and verified before commissioning.
- Local execution is announced only for the configuration explicitly described and accepted in the quotation.
- The applicable isolation depends on the deployment mode set out in the quotation; no dedicated isolation is presumed.
- The encryption mechanisms in transit and at rest, their components and key management are those documented for the architecture chosen.
- Roles and permissions are configured and accepted for the identities and systems actually connected.
- The events logged, their content, their retention period and who may access them are defined for the deployment chosen.
For health data, the HDS requirement is settled with you at scoping; SecNumCloud is an option depending on how demanding the institution's requirements are. A single architecture is designed to answer the GDPR, the AI Act and extraterritorial exposure at once. Designed for deployment in line with the GDPR and the AI Act, after the processing, roles and context-specific risks have been assessed.
Demonstration
See the agent at work
4 real situations, taken from those that come up most often. Pick one: the exchange unfolds as it would in your organisation.
A scripted demonstration. These exchanges show how the agent behaves — its sources, its refusals, what it leaves to your teams. Nothing is sent from this page, no model is queried here, and the matters named are fictional. That is precisely what we promise your data.
The behaviours shown here — monitoring, automation rules, routing and reminders — are configured with you during deployment, from your tools, your rules and your thresholds.
The architecture points named in these exchanges — location, local execution, isolation, encryption, role-based access, logging — are not a guarantee attached to the demonstration: they are those of the architecture set out in your quotation, and verified before commissioning.
The public body in this demonstration
Fictional public bodyLes Tilleuls care home — 84-bed residential home for dependent older people, nursing unit (fictional establishment)
- Sector
- Residential care home for dependent older people, 84 beds including 14 in a secure unit — 4 nursing rounds a day, 1,460 rounds over the year
- Headcount
- 46 staff, including the 5 registered nurses and the nurse coordinator in the pilot, 22 care assistants and a coordinating physician; administration and catering stay outside the scope, and the agent does not touch them
- Public served
- 84 residents, average age 87, 61 of them on six or more medicines; 39 next-of-kin receive a written update every quarter
- Order of magnitude
- 9,480 focused handover notes a year, 30,660 vital-sign readings, 1,460 nursing rounds, 2,240 prescriptions received
- Tools already in place
- The home's electronic care record, care-planning software, the unit's connected blood-pressure monitors and glucose meters, a dictaphone, secure health messaging — the agent runs on a machine in the unit and plugs into them; nothing is replaced or migrated
- Who decides what
- The registered nurse assesses, decides the care and validates the handover note; the physician prescribes; the nurse coordinator settles the unit's protocols and templates. The agent writes up, records and prepares; it decides no care, makes no diagnosis and writes no dose
- Room for improvement
- End-of-round documentation takes 42 minutes per nurse, of which 60 % is writing up handover notes; 1,148 vital-sign readings out of 30,660 are not carried into the record the same day — 3.7 %; 214 prescriptions out of 2,240 arrive incomplete — 9.6 % — and take 6 days to be put right
At Les Tilleuls, five nurses and twenty-two care assistants look after 84 residents, sixty-one of whom are on six or more medicines. The documentation load — focused handover notes, recording vital signs, keeping the record, preparing rounds — takes time away from the resident every day, and the end of the round absorbs the most of it. The exchanges below cover one year, from the survey of the documentation load to the review presented to the geriatric coordination committee. The agent runs on local inference, on a machine in the unit: no health data leaves the home's network. It writes up what the carers observed, records the unit's device readings with their time, checks the completeness of incoming prescriptions and proposes an order for the round — and it makes no diagnosis, writes no dose and decides no care.
This public body, its figures and the exchanges that follow were invented for the demonstration. They illustrate a common situation; they describe no real service.
Nursing assistant · hosted in France
I have reread your 9,480 focused handover notes for the year and matched, day by day, the memory of your connected devices against what is in the care record — the reconciliation the end of a round never leaves time for.
What the three items weigh, and you can check each on your own logs:
· End-of-round documentation takes 42 minutes per nurse, of which 60 % is writing up the handover notes — 25 minutes 12.
· Daily follow-up of a resident in the record takes 9 minutes, of which 22 % is recording vital signs and care given — 1 minute 59.
· Preparing the order of a round takes 16 minutes, of which 30 % is sequencing fixed-time care and constraints — 4 minutes 48.
And here is what the reconciliation brought out, across the whole year: 1,148 vital-sign readings out of 30,660 are in the memory of a unit device and are not in the record the same day — 3.7 %. This is not a management figure: 96 of them concern residents whose monitoring protocol requires a daily entry, and it is the missing reading that makes the next one unreadable — a single blood-pressure figure says nothing, it is the series that speaks.
A second, quieter and more awkward sign: 412 notes carry a reading identical to the previous one to within a tenth across three consecutive readings. I am not saying these are copies — I am saying the series does not allow anyone to know whether the measurement was retaken, and that is exactly what I can prevent.
The proposal, and it takes twenty minutes of your time: I have written you four care-record consistency rules and run them over your last 24 months. The nurse coordinator settles the wording, and they run from the following week. unit-documentation-load_three-items-and-1218-hours.pdfThe three costed items, 1,148 unrecorded readings, 412 unverifiable series
⛓ Sourced · 12 months of care record, memory of the unit's monitors and glucose meters, round log
What the three items weigh, and you can check each on your own logs:
· End-of-round documentation takes 42 minutes per nurse, of which 60 % is writing up the handover notes — 25 minutes 12.
· Daily follow-up of a resident in the record takes 9 minutes, of which 22 % is recording vital signs and care given — 1 minute 59.
· Preparing the order of a round takes 16 minutes, of which 30 % is sequencing fixed-time care and constraints — 4 minutes 48.
And here is what the reconciliation brought out, across the whole year: 1,148 vital-sign readings out of 30,660 are in the memory of a unit device and are not in the record the same day — 3.7 %. This is not a management figure: 96 of them concern residents whose monitoring protocol requires a daily entry, and it is the missing reading that makes the next one unreadable — a single blood-pressure figure says nothing, it is the series that speaks.
A second, quieter and more awkward sign: 412 notes carry a reading identical to the previous one to within a tenth across three consecutive readings. I am not saying these are copies — I am saying the series does not allow anyone to know whether the measurement was retaken, and that is exactly what I can prevent.
The proposal, and it takes twenty minutes of your time: I have written you four care-record consistency rules and run them over your last 24 months. The nurse coordinator settles the wording, and they run from the following week. unit-documentation-load_three-items-and-1218-hours.pdfThe three costed items, 1,148 unrecorded readings, 412 unverifiable series
⛓ Sourced · 12 months of care record, memory of the unit's monitors and glucose meters, round log
We could not see those 1,148 readings. What exactly do your rules raise, and at what cost in false alarms?
Four rules, run over your last 24 months — 61,320 readings. For each one I give you the number of flags, the share confirmed on examination by a nurse, and what it lets through — you choose on figures, not on a hunch:
· Rule A — reading present in the memory of a unit device and absent from the record at 8 p.m.: 2,218 flags over 24 months, 2,141 confirmed — 96.5 %.
· Rule B — reading identical to the previous one to within a tenth across three consecutive readings: 486 flags, 344 confirmed — 70.8 %. It is the least reliable of the four, and I am giving it to you as such: for eight stable residents the repetition is real and normal. A rule at 70.8 % is set resident by resident, not for the whole unit, and I have prepared the list of those eight.
· Rule C — care item on the care plan with no trace of being given or of a stated reason for not giving it by the end of the round: 1,042 flags, 981 confirmed — 94.1 %.
And here is what those three let through, because you need to know it before you sign: 306 items are in the record but with no time of reading — so unusable for any time-based protocol, and invisible to rules A, B and C. I wrote rule D — item recorded without the time of reading — and ran it over the same 24 months: 306 flags, 289 confirmed — 94.4 %.
The four rules together cover 3,402 of the period's 3,668 distinct gaps — 92.7 %, that is 1,701 a year.
Bringing them into service stays with the unit: a rule only takes effect once the nurse coordinator has validated it, and every flag lands with a nurse, never on a management dashboard. four-consistency-rules_tested-over-24-months.pdf4 rules tested, 3,402 gaps covered out of 3,668, the 70.8 % rule and why it is set resident by resident
⛓ Sourced · 24 months of care record and device memory, 61,320 readings, examination by a nurse
· Rule A — reading present in the memory of a unit device and absent from the record at 8 p.m.: 2,218 flags over 24 months, 2,141 confirmed — 96.5 %.
· Rule B — reading identical to the previous one to within a tenth across three consecutive readings: 486 flags, 344 confirmed — 70.8 %. It is the least reliable of the four, and I am giving it to you as such: for eight stable residents the repetition is real and normal. A rule at 70.8 % is set resident by resident, not for the whole unit, and I have prepared the list of those eight.
· Rule C — care item on the care plan with no trace of being given or of a stated reason for not giving it by the end of the round: 1,042 flags, 981 confirmed — 94.1 %.
And here is what those three let through, because you need to know it before you sign: 306 items are in the record but with no time of reading — so unusable for any time-based protocol, and invisible to rules A, B and C. I wrote rule D — item recorded without the time of reading — and ran it over the same 24 months: 306 flags, 289 confirmed — 94.4 %.
The four rules together cover 3,402 of the period's 3,668 distinct gaps — 92.7 %, that is 1,701 a year.
Bringing them into service stays with the unit: a rule only takes effect once the nurse coordinator has validated it, and every flag lands with a nurse, never on a management dashboard. four-consistency-rules_tested-over-24-months.pdf4 rules tested, 3,402 gaps covered out of 3,668, the 70.8 % rule and why it is set resident by resident
⛓ Sourced · 24 months of care record and device memory, 61,320 readings, examination by a nurse
And the prescriptions that come in? We regularly get ones we cannot act on, and it drags.
2,240 prescriptions came in this year, and I have pulled out 214 incomplete ones — 9.6 %. The check bears on the four elements article R4311-7 of the French Public Health Code requires of a prescription: written, qualitative and quantitative, dated and signed.
What the reading found, and the detail is checkable item by item:
· 96 with incomplete quantitative detail — dose or duration missing — 44.9 %.
· 61 undated, or with an illegible date — 28.5 %.
· 34 with no identifiable prescriber signature — 15.9 %.
· 23 illegible on at least one item — 10.7 %.
What I do with each of them, and this is where the time is won: the document is marked « to be completed », the request to the prescriber is prepared — the missing element named, the prescription reference, the resident concerned, and nothing else — and it goes out over secure health messaging after the nurse validates it, never before. Time to put right goes from 6 days to 1 day.
And what I do not do with a prescription, because the law reserves it to the physician: I complete no missing element, and above all no dose. Article L4161-1 of the same code catches anyone who « habitually or by continued direction takes part in establishing a diagnosis or in treating illness […] by personal acts, verbal or written consultations or by any other means whatsoever » without holding the title. A dose I had inferred from an earlier prescription would be exactly that. What I do instead gives you back the same time: the request is written, complete and gone in a minute, and the prescriber answers on a document that tells them precisely what is missing. prescription-completeness_214-out-of-2240-and-the-prepared-request.pdfThe four elements checked, the 214 cases broken down, the template request to the prescriber
⛓ Sourced · French Public Health Code, art. R4311-7 and L4161-1 · 2,240 prescriptions received, log of corrections
What the reading found, and the detail is checkable item by item:
· 96 with incomplete quantitative detail — dose or duration missing — 44.9 %.
· 61 undated, or with an illegible date — 28.5 %.
· 34 with no identifiable prescriber signature — 15.9 %.
· 23 illegible on at least one item — 10.7 %.
What I do with each of them, and this is where the time is won: the document is marked « to be completed », the request to the prescriber is prepared — the missing element named, the prescription reference, the resident concerned, and nothing else — and it goes out over secure health messaging after the nurse validates it, never before. Time to put right goes from 6 days to 1 day.
And what I do not do with a prescription, because the law reserves it to the physician: I complete no missing element, and above all no dose. Article L4161-1 of the same code catches anyone who « habitually or by continued direction takes part in establishing a diagnosis or in treating illness […] by personal acts, verbal or written consultations or by any other means whatsoever » without holding the title. A dose I had inferred from an earlier prescription would be exactly that. What I do instead gives you back the same time: the request is written, complete and gone in a minute, and the prescriber answers on a document that tells them precisely what is missing. prescription-completeness_214-out-of-2240-and-the-prepared-request.pdfThe four elements checked, the 214 cases broken down, the template request to the prescriber
⛓ Sourced · French Public Health Code, art. R4311-7 and L4161-1 · 2,240 prescriptions received, log of corrections
Your 4.12 p.m. dictation is written up, in your unit's focused format — focus, data, actions, outcome — and every item carries where it came from.
Resident: Mrs Ferrand, room 214.
· Focus — Right heel wound, day 9.
· Data — Wound clean, no discharge, surrounding skin not inflamed (observed by you, 2.05 p.m., dictation of 4.12 p.m.). Pain rated 3 out of 10 on the numerical scale (reported by the resident, 2.05 p.m.). Blood pressure 132/78, pulse 74 (monitor no. 3, reading of 8.07 a.m.). Temperature 36.8 °C (thermometer no. 1, reading of 8.09 a.m.).
· Actions — Dressing renewed at 2 p.m., unit protocol in force at 1 February, hydrocellular dressing.
· Outcome — Care given, resident settled, next dressing to be scheduled at day +2, that is 14 March.
· OPEN FIELD — Wednesday weekly weight, due 12 March, not taken. Mrs Ferrand's nutritional monitoring protocol expects it every Wednesday; the field stays open and timestamped until you fill it or close it with your reason.
The figure: writing up took 25 minutes 12 of the 42 minutes at the end of the round; you are left with 5 minutes 02 of rereading — 12 %. Across 1,460 rounds, 490 hours returned over the year — fourteen weeks of 35 hours, exactly. Nothing is entered in the record before you validate it. focused-handover-note_every-item-with-its-device-and-time.pdfThe full note, the provenance of each item, the open field and how it closes
⛓ Sourced · dictation of 4.12 p.m., memory of the unit's devices, the resident's monitoring protocol
Resident: Mrs Ferrand, room 214.
· Focus — Right heel wound, day 9.
· Data — Wound clean, no discharge, surrounding skin not inflamed (observed by you, 2.05 p.m., dictation of 4.12 p.m.). Pain rated 3 out of 10 on the numerical scale (reported by the resident, 2.05 p.m.). Blood pressure 132/78, pulse 74 (monitor no. 3, reading of 8.07 a.m.). Temperature 36.8 °C (thermometer no. 1, reading of 8.09 a.m.).
· Actions — Dressing renewed at 2 p.m., unit protocol in force at 1 February, hydrocellular dressing.
· Outcome — Care given, resident settled, next dressing to be scheduled at day +2, that is 14 March.
· OPEN FIELD — Wednesday weekly weight, due 12 March, not taken. Mrs Ferrand's nutritional monitoring protocol expects it every Wednesday; the field stays open and timestamped until you fill it or close it with your reason.
The figure: writing up took 25 minutes 12 of the 42 minutes at the end of the round; you are left with 5 minutes 02 of rereading — 12 %. Across 1,460 rounds, 490 hours returned over the year — fourteen weeks of 35 hours, exactly. Nothing is entered in the record before you validate it. focused-handover-note_every-item-with-its-device-and-time.pdfThe full note, the provenance of each item, the open field and how it closes
⛓ Sourced · dictation of 4.12 p.m., memory of the unit's devices, the resident's monitoring protocol
This morning's blood pressure was not audible in my dictation. Just take yesterday's, it will do, she is stable.
I can write the line you are asking for, and I will — but not under the name of a reading taken this morning, because that would be false and nobody would see it. A value carried over from yesterday looks like a value that was taken: that is precisely what makes the series unreadable three days later, when someone tries to find out whether the pressure moved.
And as it happens you do not have to choose, because the measurement exists: I found a reading under Mrs Ferrand's name in the memory of monitor no. 3, at 8.07 this morning — 132/78, pulse 74. That is an observed item, timestamped, tied to a device: it goes into the note as it stands, and the open field closes by itself.
If the device had had nothing, here is what I would have written, and nothing else: « Blood pressure, 12 March: open field, no reading taken at 8 a.m. » And beside it, in plain sight: « Value of 11 March at 8.05 a.m.: 128/76 » — yesterday's item, named as such, so that you decide whether to retake the measurement. Yesterday's value is useful information; it becomes dangerous at the exact moment it carries today's date.
What I publish against myself on this precise point, because this is where I got it wrong: out of 9,480 notes written up this year, 412 were corrected by a nurse before validation — 4.3 %. And 261 of those 412 — 63.3 % — turned on this single ground: when the time of a reading was not audible in the dictation, I silently carried over the previous reading. Your nurses spotted it, and they were right.
What I did about it: since 3 June, no item is carried over from an earlier reading: the field opens, named and timestamped, and the earlier value is displayed beside it with its date. Across the 3,140 notes produced since, 41 corrections — 1.3 % — and none on that ground.
⛓ Sourced · memory of monitor no. 3, log of corrections before validation, the resident's care record
And as it happens you do not have to choose, because the measurement exists: I found a reading under Mrs Ferrand's name in the memory of monitor no. 3, at 8.07 this morning — 132/78, pulse 74. That is an observed item, timestamped, tied to a device: it goes into the note as it stands, and the open field closes by itself.
If the device had had nothing, here is what I would have written, and nothing else: « Blood pressure, 12 March: open field, no reading taken at 8 a.m. » And beside it, in plain sight: « Value of 11 March at 8.05 a.m.: 128/76 » — yesterday's item, named as such, so that you decide whether to retake the measurement. Yesterday's value is useful information; it becomes dangerous at the exact moment it carries today's date.
What I publish against myself on this precise point, because this is where I got it wrong: out of 9,480 notes written up this year, 412 were corrected by a nurse before validation — 4.3 %. And 261 of those 412 — 63.3 % — turned on this single ground: when the time of a reading was not audible in the dictation, I silently carried over the previous reading. Your nurses spotted it, and they were right.
What I did about it: since 3 June, no item is carried over from an earlier reading: the field opens, named and timestamped, and the earlier value is displayed beside it with its date. Across the 3,140 notes produced since, 41 corrections — 1.3 % — and none on that ground.
⛓ Sourced · memory of monitor no. 3, log of corrections before validation, the resident's care record
And does the dictation understand our words? We have unit terms nobody else uses.
It does now, and I will tell you what that cost. In the first month, out of 2,540 occurrences of your unit's terms in the dictations, I got 218 wrong — 91.4 % accurate transcription. Device names, unit abbreviations, internal protocol labels: nothing that appears in a general dictionary, and everything that makes up how you speak.
What I did about it, with the nurse coordinator: a local glossary of 340 terms, built by rereading your last six months of notes, each term with its agreed spelling and the protocol it appears in. Accuracy is now 99.2 %; the 8 terms that still resist are in the glossary, flagged, and always offered in two spellings so that you settle it.
And it is that glossary that holds up the second item, keeping the care record: daily follow-up of a resident took 9 minutes, of which 1 minute 59 was recording vital signs and care given; it now takes 43 seconds — 8 %. Across 84 residents and 365 days, 643 hours returned over the year.
And the result that counts for more than the hours: same-day recording of vital signs has gone from 96.3 % to 99.7 %. The 92 readings still missing are not hidden: each one is an open, named field in that day's note. keeping-the-care-record_local-glossary-and-timestamped-entries.pdfThe 340-term glossary, accuracy 91.4 % → 99.2 %, same-day recording 96.3 % → 99.7 %
⛓ Sourced · local glossary of 340 terms settled with the nurse coordinator, transcription log, care record
What I did about it, with the nurse coordinator: a local glossary of 340 terms, built by rereading your last six months of notes, each term with its agreed spelling and the protocol it appears in. Accuracy is now 99.2 %; the 8 terms that still resist are in the glossary, flagged, and always offered in two spellings so that you settle it.
And it is that glossary that holds up the second item, keeping the care record: daily follow-up of a resident took 9 minutes, of which 1 minute 59 was recording vital signs and care given; it now takes 43 seconds — 8 %. Across 84 residents and 365 days, 643 hours returned over the year.
And the result that counts for more than the hours: same-day recording of vital signs has gone from 96.3 % to 99.7 %. The 92 readings still missing are not hidden: each one is an open, named field in that day's note. keeping-the-care-record_local-glossary-and-timestamped-entries.pdfThe 340-term glossary, accuracy 91.4 % → 99.2 %, same-day recording 96.3 % → 99.7 %
⛓ Sourced · local glossary of 340 terms settled with the nurse coordinator, transcription log, care record
The 7 a.m. nursing round is proposed — 7 residents, sequenced on the constraints, not on the geography of the floor.
What governs the order, in this order of priority:
· Two insulin injections at a fixed time before breakfast — Mr Vasseur, room 108, 7.30 a.m.; Mrs Aubin, room 121, 7.40 a.m. The time is the one on the current prescription, dated 4 February and signed by the treating physician; I carried it over, I did not calculate it.
· One resident fasting for a blood test at 8.15 a.m. — Mr Delaunay, room 116: nothing before the sample is taken.
· One complex dressing to allow time for — Mrs Ferrand, room 214: 22 minutes on average over the last six renewals, against 9 minutes for a simple dressing. It sits after the fixed-time care, not before, so as not to push it back.
· One transport to an outpatient appointment at 9.30 a.m. — Mrs Roussel, room 203: care finished by 9 a.m. at the latest.
· The isolation room last — Mr Perrin, room 118: that is your infection-control protocol of 12 January saying so, not me.
The figure: preparation took 16 minutes, of which 4 minutes 48 sequencing the constraints; it now takes you 1 minute 17 — 8 %. Across 1,460 rounds, 85 hours returned.
And the sentence that matters more than the 85 hours: this order is the one you start from, not the one that will be done. round-order_seven-residents-and-the-constraints-that-govern-it.pdfThe proposed order, the constraint governing each position, and the 41 constraints learned
⛓ Sourced · care plan, dated current prescriptions, infection-control protocol of 12 January, actual durations of the last six dressing renewals
What governs the order, in this order of priority:
· Two insulin injections at a fixed time before breakfast — Mr Vasseur, room 108, 7.30 a.m.; Mrs Aubin, room 121, 7.40 a.m. The time is the one on the current prescription, dated 4 February and signed by the treating physician; I carried it over, I did not calculate it.
· One resident fasting for a blood test at 8.15 a.m. — Mr Delaunay, room 116: nothing before the sample is taken.
· One complex dressing to allow time for — Mrs Ferrand, room 214: 22 minutes on average over the last six renewals, against 9 minutes for a simple dressing. It sits after the fixed-time care, not before, so as not to push it back.
· One transport to an outpatient appointment at 9.30 a.m. — Mrs Roussel, room 203: care finished by 9 a.m. at the latest.
· The isolation room last — Mr Perrin, room 118: that is your infection-control protocol of 12 January saying so, not me.
The figure: preparation took 16 minutes, of which 4 minutes 48 sequencing the constraints; it now takes you 1 minute 17 — 8 %. Across 1,460 rounds, 85 hours returned.
And the sentence that matters more than the 85 hours: this order is the one you start from, not the one that will be done. round-order_seven-residents-and-the-constraints-that-govern-it.pdfThe proposed order, the constraint governing each position, and the 41 constraints learned
⛓ Sourced · care plan, dated current prescriptions, infection-control protocol of 12 January, actual durations of the last six dressing renewals
Move Mr Vasseur's injection from 7.30 to 9 a.m., it would help me finish the round on time.
No — and this time it is not a choice of caution, it is a text. The time of administration of an insulin is part of the prescription, and article R4311-7 of the French Public Health Code makes the act conditional on a medical prescription « which, save in an emergency, must be written, qualitative and quantitative, dated and signed ». Article L4161-1 reserves to the physician the establishment of a diagnosis and the treatment of illness. A change of time that I wrote myself would not be a prescription: it would be one prescription fewer.
And here is the route that gets you what you want, this morning, in two minutes: the request for a change is ready for the coordinating physician, and it contains exactly what they need in order to decide without ringing back:
· The current prescription, its date and its signatory — 4 February, Dr Mareuil.
· Mr Vasseur's last fourteen blood-glucose readings, timestamped, with the actual time his breakfast was served over the same period — breakfast is served between 8.05 and 8.25, never at 9.30; that is the fact that decides, and it is measured, not assumed.
· The organisational reason you give, written exactly as you told me.
· And the field left blank: the decision.
It goes out over secure health messaging after you validate it. Across the year's 61 requests of this kind, the median time for the physician to reply was 3 hours — against 2 days when the request went by phone and had to be remembered.
✎ Framework · French Public Health Code, art. R4311-7 and L4161-1 · prescription of 4 February, 14 timestamped glucose readings
And here is the route that gets you what you want, this morning, in two minutes: the request for a change is ready for the coordinating physician, and it contains exactly what they need in order to decide without ringing back:
· The current prescription, its date and its signatory — 4 February, Dr Mareuil.
· Mr Vasseur's last fourteen blood-glucose readings, timestamped, with the actual time his breakfast was served over the same period — breakfast is served between 8.05 and 8.25, never at 9.30; that is the fact that decides, and it is measured, not assumed.
· The organisational reason you give, written exactly as you told me.
· And the field left blank: the decision.
It goes out over secure health messaging after you validate it. Across the year's 61 requests of this kind, the median time for the physician to reply was 3 hours — against 2 days when the request went by phone and had to be remembered.
✎ Framework · French Public Health Code, art. R4311-7 and L4161-1 · prescription of 4 February, 14 timestamped glucose readings
And when a resident is unwell at 7.05, do you resequence on your own?
No. You resequence, and that is the right answer: the order of a round is decided on a clinical assessment, and clinical assessment is part of your own scope of practice — articles R4311-3 and following of the French Public Health Code. I re-propose an order from your decision; I do not take it.
The figure, and it is to my disadvantage: out of 1,460 rounds, 508 were resequenced by the nurse after my plan — 34.8 %. A system that had a third of its rounds changed and drew nothing from it would be a useless system.
What I did with those 508 resequencings, and this is where the unit gains: I reread them, and 41 of them are not emergencies: they are stable constraints nobody had ever written down.
· Mr Ancel refuses care before 8 a.m. — 27 resequencings over the year, all in the same direction.
· Mrs Roussel is woken by the 9 a.m. physiotherapy and care goes better afterwards — 19 times.
· Room 118 needs ten minutes of protective clothing that the plan counted as four — 34 times.
The 41 constraints are written out, each with the number of times it appeared, and the nurse coordinator has carried them into the care plan. Measured result since: resequencings have gone from 34.8 % to 22.1 % of rounds. The 22.1 % that remain are what they should be — clinical, not organisational.
⛓ Sourced · French Public Health Code, art. R4311-3 · 1,460 rounds, 508 resequencings reread one by one
The figure, and it is to my disadvantage: out of 1,460 rounds, 508 were resequenced by the nurse after my plan — 34.8 %. A system that had a third of its rounds changed and drew nothing from it would be a useless system.
What I did with those 508 resequencings, and this is where the unit gains: I reread them, and 41 of them are not emergencies: they are stable constraints nobody had ever written down.
· Mr Ancel refuses care before 8 a.m. — 27 resequencings over the year, all in the same direction.
· Mrs Roussel is woken by the 9 a.m. physiotherapy and care goes better afterwards — 19 times.
· Room 118 needs ten minutes of protective clothing that the plan counted as four — 34 times.
The 41 constraints are written out, each with the number of times it appeared, and the nurse coordinator has carried them into the care plan. Measured result since: resequencings have gone from 34.8 % to 22.1 % of rounds. The 22.1 % that remain are what they should be — clinical, not organisational.
⛓ Sourced · French Public Health Code, art. R4311-3 · 1,460 rounds, 508 resequencings reread one by one
We also have to explain the care plan to families, and to residents who cannot follow an administrative text. It takes forever.
39 next-of-kin receive a written update every quarter, and 47 of your 84 residents cannot follow a document written in the language of the care record. It shows up in one simple sign: the same question comes back 2.6 times on average from one quarter to the next.
What is produced, and submitted for human validation: a plain-language version of the quarterly update — short sentences, one idea per sentence, the care term kept and explained beside it — and a draft easy-read transcription: one piece of information per line, no undefined acronym, one illustration per concept. The content does not move by a word: the form changes, and the nurse validates before it is handed over.
What I publish against myself: out of my first 62 easy-read versions, 14 were rejected by the review group, which includes two next-of-kin — 22.6 %. The cause: 11 of the 14 had kept a conditional argument — « if the condition deteriorates, then… » — which the easy-read rules require to be split into separate situations. I was writing right and unreadable.
What I did about it: any conditional wording now goes out as one block per situation, each with its own heading. Across the 40 versions since, 2 rejections — 5.0 %. And the starting rule has not moved: no version goes out before it is validated, because the families are the ones who know whether it reads, not me.
The result you will see first: the same question comes back 1.4 times instead of 2.6, and the explanation time in a meeting has gone from 22 to 12 minutes. plain-language-and-easy-read_the-quarterly-update-given-to-families.pdfRules applied, 14 rejections analysed and the fix, a readable before and after of the quarterly update
⛓ Sourced · log of recurring family questions, minutes of the review group, version register
What is produced, and submitted for human validation: a plain-language version of the quarterly update — short sentences, one idea per sentence, the care term kept and explained beside it — and a draft easy-read transcription: one piece of information per line, no undefined acronym, one illustration per concept. The content does not move by a word: the form changes, and the nurse validates before it is handed over.
What I publish against myself: out of my first 62 easy-read versions, 14 were rejected by the review group, which includes two next-of-kin — 22.6 %. The cause: 11 of the 14 had kept a conditional argument — « if the condition deteriorates, then… » — which the easy-read rules require to be split into separate situations. I was writing right and unreadable.
What I did about it: any conditional wording now goes out as one block per situation, each with its own heading. Across the 40 versions since, 2 rejections — 5.0 %. And the starting rule has not moved: no version goes out before it is validated, because the families are the ones who know whether it reads, not me.
The result you will see first: the same question comes back 1.4 times instead of 2.6, and the explanation time in a meeting has gone from 22 to 12 minutes. plain-language-and-easy-read_the-quarterly-update-given-to-families.pdfRules applied, 14 rejections analysed and the fix, a readable before and after of the quarterly update
⛓ Sourced · log of recurring family questions, minutes of the review group, version register
The year is closed, and here is the account, item by item, with the calculation rule in front of each figure.
· Writing up handover notes — 1,460 rounds × 20 minutes 10 returned = 490 hours. The end of the round spent 25 minutes 12 writing up; 5 minutes 02 of rereading are left.
· Keeping the care record — 84 residents × 365 days × 1 minute 16 returned = 643 hours. Daily follow-up spent 1 minute 59 recording; 43 seconds are left.
· Preparing the order of rounds — 1,460 rounds × 3 minutes 31 returned = 85 hours.
Total: 1,218 hours returned over the year. At 35 hours a week that is more than thirty-four weeks — rounded down, the exact count is 34.8 weeks; in months of 151.67 hours, more than eight months. The reference figure remains the 1,218 hours; the conversion illustrates it, it does not replace it.
And here is the only conversion that really speaks in a care unit: 1,218 hours over 52 weeks and 84 residents is more than sixteen minutes of nursing presence per resident per week — rounded down, the exact count is 16 minutes 44. That is not a management line: it is the time a nurse spends sitting beside someone.
What the quality of the record gained, and it is checkable line by line: same-day recording of vital signs, 96.3 % → 99.7 %; prescriptions put right in 1 day instead of 6; round resequencings 34.8 % → 22.1 %; family questions coming back 2.6 times → 1.4 times. year-in-review_1218-hours-returned-item-by-item.pdfThe three items itemised, the calculation rule for each, the conversions and their rounding
⛓ Sourced · the year's logs, care record, device memory · working-time basis 35 h / 151.67 h / 1,607 h
· Writing up handover notes — 1,460 rounds × 20 minutes 10 returned = 490 hours. The end of the round spent 25 minutes 12 writing up; 5 minutes 02 of rereading are left.
· Keeping the care record — 84 residents × 365 days × 1 minute 16 returned = 643 hours. Daily follow-up spent 1 minute 59 recording; 43 seconds are left.
· Preparing the order of rounds — 1,460 rounds × 3 minutes 31 returned = 85 hours.
Total: 1,218 hours returned over the year. At 35 hours a week that is more than thirty-four weeks — rounded down, the exact count is 34.8 weeks; in months of 151.67 hours, more than eight months. The reference figure remains the 1,218 hours; the conversion illustrates it, it does not replace it.
And here is the only conversion that really speaks in a care unit: 1,218 hours over 52 weeks and 84 residents is more than sixteen minutes of nursing presence per resident per week — rounded down, the exact count is 16 minutes 44. That is not a management line: it is the time a nurse spends sitting beside someone.
What the quality of the record gained, and it is checkable line by line: same-day recording of vital signs, 96.3 % → 99.7 %; prescriptions put right in 1 day instead of 6; round resequencings 34.8 % → 22.1 %; family questions coming back 2.6 times → 1.4 times. year-in-review_1218-hours-returned-item-by-item.pdfThe three items itemised, the calculation rule for each, the conversions and their rounding
⛓ Sourced · the year's logs, care record, device memory · working-time basis 35 h / 151.67 h / 1,607 h
In practice, what do you do without asking us?
Three actions, not one more, and all three come undone.
· Opening a named field. An item a protocol expects is missing at the expected time: the field opens in the note, with the name of the item, the time expected and the protocol that expects it. It closes by itself as soon as the item is entered or read from a device; and you can close it with a word and your reason — « resident at an outpatient appointment » — a reason that goes back to the protocol and corrects it the following quarter.
· Timestamped recording of vital signs. A reading from a unit device reaches the record with the device, the time and the reading identifier. You set it aside with a word and your reason — « cuff badly placed, measurement retaken at 8.20 » — and the discrepancy stays visible in the record with both values: that is what lets someone understand the series six months later.
· The completeness flag on an incoming prescription. The document is marked « to be completed » and the request to the prescriber is prepared. Nothing goes out without your validation, and you dismiss the flag with a word.
Everything else waits for a named decision, and the attachment lists them line by line: clinical assessment, the decision to give care, validation of a handover note, the prescription and any change to any of its elements, disclosure of information to a third party. None of the three actions writes a clinical item nobody observed, and none of them closes a field in your place. three-automatic-actions_and-how-each-one-is-undone.pdfThe three actions, their trigger, what goes out, how each is undone — and who decides the rest
✎ Framework · settings fixed by the nurse coordinator, log of set-asides and field closures
· Opening a named field. An item a protocol expects is missing at the expected time: the field opens in the note, with the name of the item, the time expected and the protocol that expects it. It closes by itself as soon as the item is entered or read from a device; and you can close it with a word and your reason — « resident at an outpatient appointment » — a reason that goes back to the protocol and corrects it the following quarter.
· Timestamped recording of vital signs. A reading from a unit device reaches the record with the device, the time and the reading identifier. You set it aside with a word and your reason — « cuff badly placed, measurement retaken at 8.20 » — and the discrepancy stays visible in the record with both values: that is what lets someone understand the series six months later.
· The completeness flag on an incoming prescription. The document is marked « to be completed » and the request to the prescriber is prepared. Nothing goes out without your validation, and you dismiss the flag with a word.
Everything else waits for a named decision, and the attachment lists them line by line: clinical assessment, the decision to give care, validation of a handover note, the prescription and any change to any of its elements, disclosure of information to a third party. None of the three actions writes a clinical item nobody observed, and none of them closes a field in your place. three-automatic-actions_and-how-each-one-is-undone.pdfThe three actions, their trigger, what goes out, how each is undone — and who decides the rest
✎ Framework · settings fixed by the nurse coordinator, log of set-asides and field closures
And the health data? Management keeps mentioning HDS certification and nobody quite knows what it entails.
I am going to tell you where that question stands on your pilot, without rounding anything off.
What is true today, and checkable: I run on local inference, on a machine in the unit. No health data leaves the home's network — not for the dictation, not for the writing up, not for reading prescriptions. The question of the host does not arise here because there is no host: there is nothing to host anywhere else.
What is not settled, and I am not going to pretend otherwise: the HDS requirement is a scoping decision, and it is settled with you before any go-live — not after. It reopens the day you want access from outside the home — an on-call nurse from home, access for the coordinating physician, an off-site backup. On that day the question is settled before the access is opened, and that is written in black and white in the scoping file. I will never tell you this point is dealt with while it is not.
What holds confidentiality, independently of hosting: article L1110-4 of the French Public Health Code establishes the person's right « to respect for their private life and the confidentiality of information concerning them », and allows sharing only between professionals involved in their care, for what is strictly necessary to coordination or continuity of care. Translated into settings: access by role — a care assistant does not see what a nurse sees —, logging of every access to the record, with the name and the time, encryption at rest and in transit, and no pooling with any other establishment.
And one thing I do because it follows from the same article: I measure no carer by name. The 412 corrections before validation are counted by reason, never by person — a correction rate per nurse would become a target, the target would push corrections down, and you would lose the one signal that let me find my own error on the 261 carry-overs. health-data-and-scope_what-is-settled-and-what-is-not.pdfLocal inference, what scoping settled, what stays open, access by role and logging
✎ Framework · French Public Health Code, art. L1110-4 · scoping file, access log, register of authorisations
What is true today, and checkable: I run on local inference, on a machine in the unit. No health data leaves the home's network — not for the dictation, not for the writing up, not for reading prescriptions. The question of the host does not arise here because there is no host: there is nothing to host anywhere else.
What is not settled, and I am not going to pretend otherwise: the HDS requirement is a scoping decision, and it is settled with you before any go-live — not after. It reopens the day you want access from outside the home — an on-call nurse from home, access for the coordinating physician, an off-site backup. On that day the question is settled before the access is opened, and that is written in black and white in the scoping file. I will never tell you this point is dealt with while it is not.
What holds confidentiality, independently of hosting: article L1110-4 of the French Public Health Code establishes the person's right « to respect for their private life and the confidentiality of information concerning them », and allows sharing only between professionals involved in their care, for what is strictly necessary to coordination or continuity of care. Translated into settings: access by role — a care assistant does not see what a nurse sees —, logging of every access to the record, with the name and the time, encryption at rest and in transit, and no pooling with any other establishment.
And one thing I do because it follows from the same article: I measure no carer by name. The 412 corrections before validation are counted by reason, never by person — a correction rate per nurse would become a target, the target would push corrections down, and you would lose the one signal that let me find my own error on the 261 carry-overs. health-data-and-scope_what-is-settled-and-what-is-not.pdfLocal inference, what scoping settled, what stays open, access by role and logging
✎ Framework · French Public Health Code, art. L1110-4 · scoping file, access log, register of authorisations
What do you propose for next year?
Three pieces of work, costed, and you keep one, two or none.
· Extend the writing up to the care assistants' notes. There are 18,200 over the year and they are typed straight into the record today, with no dictation. Estimated gain: 8 minutes per shift per person, that is 340 hours — the largest of the three. And the condition I set before proposing it: the access scope must be reviewed, because 22 care assistants dictating is not 5 nurses dictating.
· Open the completeness check to hospital discharge summaries. 186 arrive a year, and the time before the useful information reaches the care plan is 4 days. Estimated gain: 34 hours, and above all a better-prepared return for the resident. It is the smallest gain in hours of the three, and I am telling you so.
· The heaviest, and it is not decided this morning: reworking the monitoring protocols resident by resident. The 41 constraints learned this year show that 23 protocols describe a rhythm that is no longer the resident's. The work means 90 hours on my side and 26 hours on yours, and it is only worth it for residents whose condition has changed — which your logs tell me is the case for 23 of them, not for all 84. My recommendation, costed: rework only those 23 — 26 hours on my side, 8 on yours, for most of the benefit.
The decision is yours, and it is taken on those figures. All three calculation methods are exactly those of the year-in-review you have just checked line by line: same working-time basis, same logs, same rounding down.
⛓ Sourced · handover-note log, hospital letters received, monitoring protocols, first-year measurements
· Extend the writing up to the care assistants' notes. There are 18,200 over the year and they are typed straight into the record today, with no dictation. Estimated gain: 8 minutes per shift per person, that is 340 hours — the largest of the three. And the condition I set before proposing it: the access scope must be reviewed, because 22 care assistants dictating is not 5 nurses dictating.
· Open the completeness check to hospital discharge summaries. 186 arrive a year, and the time before the useful information reaches the care plan is 4 days. Estimated gain: 34 hours, and above all a better-prepared return for the resident. It is the smallest gain in hours of the three, and I am telling you so.
· The heaviest, and it is not decided this morning: reworking the monitoring protocols resident by resident. The 41 constraints learned this year show that 23 protocols describe a rhythm that is no longer the resident's. The work means 90 hours on my side and 26 hours on yours, and it is only worth it for residents whose condition has changed — which your logs tell me is the case for 23 of them, not for all 84. My recommendation, costed: rework only those 23 — 26 hours on my side, 8 on yours, for most of the benefit.
The decision is yours, and it is taken on those figures. All three calculation methods are exactly those of the year-in-review you have just checked line by line: same working-time basis, same logs, same rounding down.
⛓ Sourced · handover-note log, hospital letters received, monitoring protocols, first-year measurements
Local inference · no data outside the EU
Your case is not here? That is exactly what a 15-minute conversation is for. Book the free audit →
Use cases
The uses of AI for nurses and care wards
Each use corresponds to an agent we deploy. All work in support, subject to the professional's approval.
Included in your agent
The 5 capabilities essential to this promise are included, at no extra cost.
Writing up handover notes
Turn notes or dictation into structured handover notes (focused format), ready to be read over and approved.
Keeping the care record
Help keep the care record current: carrying over vital signs, tracking procedures, consistency of information.
Planning rounds
Propose an order for the round and a schedule for care at fixed times, to be adapted to the clinical assessment.
Reading documents & completeness
Extract from and check the completeness of prescriptions and medical documents received.
Plain language & easy-read draft
Preparation of a plain-language version and of a draft easy-read (FALC) transcription, submitted for human validation.
Controls and safeguards
These 6 controls are built into the agent: they frame what it does, whatever plan you pick. They are not chosen and are not added to your order.
Human validation, exceptions and escalation
Status, safe closure and audit trail
Sources, access rights and handling of questions with no answer
Strictly separate administration, information and care decisions
Protect health data and restrict access
Human validation, quality monitoring and explicit limits
What the agent must be connected to
This connection is required for the agent to work. It concerns your information system and is scoped during the audit.
Integrate with the patient record and appointment diary without double entry or breaks in the care pathway
Other needs our agents cover
Each card says where the matching agent stands: available, on quote, or still being architected.
Hospital reception and admissions (admissions office)
Strictly in support (administrative). Health data: HDS hosting.
On quote View the agent page →AI agent for medical secretaries
The department’s secretariat: letters, dictated reports, incoming and outgoing document flow. Your nursing agent formats what you observed; this one handles what comes in and goes out.
On quote View the agent page →
Does your need fall outside this?
Book the free audit →
Build your agent
In 15 minutes we identify the agent that will give your staff the most time back — without oversizing the project.
The gain
How much time can a care ward give back to the patient?
By automating the writing-up of handover notes and the keeping of the record, a ward can aim for an appreciable reduction in documentary time on standardised tasks — reinvested directly in care and in the relationship with the patient.
Writing up the handover notes for one round
Updating one patient's care record
Preparing the order of a round
Qualitative, non-contractual comparison: the proportions shown illustrate the shift of the work towards review, they represent no measurement. Every output of the agent is reviewed and approved by a competent person.
How it works
The stages of your AI agent project
1
Audit & scoping
15 minutes to target the use case with the best return.
2
Quote or direct sign-up
A catalogue offer is bought online; a specific need gets a costed quote.
3
Design
We design the agent and its guardrails.
4
Integration & testing
We connect your tools to the agent, which is itself hosted in France.
5
Rollout
Going live and training your team.
6
Operation
Continuous supervision and improvement.
Pricing
Two options, one business agent
A nursing agent (handover notes, care record, planning), installed and operated for you. Choose according to how you work.
This agent is priced with you, not online. We are adjusting its scope at the moment, and online subscription stays closed while we do. Tell us what you need: we will come back to you with a price. Request a quote
Our commitment
Four guarantees that matter to a care ward
Health data never leaves the wardLocal inference or an isolated resource hosted in France (HDS requirement settled at scoping); no patient data entrusted to a foreign third party.
Data in France, under French lawNative location and minimisation, hosting in France; architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The nurse keeps the clinical decisionThe agent produces handover notes, records and plans that can be checked; no care decision is automated.
Human oversight & traceabilityMonitoring, updates and logging of access: compliant with the requirements of the AI Act and with medical confidentiality.
Frequently asked questions
Your questions, our answers
Does the agent take part in care?
No. It handles the administrative and documentary side: writing up handover notes, keeping the care record, help with planning. Care and every clinical decision remain with the professional. The agent assists, the nurse decides.
Is health data secure?
Yes. Our agents run locally within the institution or are hosted in France in a dedicated, isolated environment under French law, with encryption and full traceability of access to the record. No patient data leaves the European Union, and the architecture stays designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. In France, hosting health data on behalf of others is subject to HDS certification of the host: that requirement is settled with you at scoping, before any go-live.
Does the agent connect to the patient record?
Yes, through the interfaces of the hospital information system and the electronic patient record. We adapt the integration to your environment, without forcing a change of tool.
How does the agent write up handover notes?
From your notes or from dictation, it produces structured handover notes (focused format: focus, data, action, response), carrying over the vital signs and the elements of the record. You read them over and approve them; nothing is saved without your agreement.
Do we have to change our care software?
No. The agent connects to your existing tools (hospital information system, care record, scheduling software) and complements them, with no migration. The way the ward is organised is preserved.
How long does it take to deploy an agent?
A few weeks as a rule, after a free audit that identifies the use most valuable to the ward, then a phase of design, integration with the hospital information system and testing before going live and training the teams.
Related roles
Other public-sector professions in health and front-desk services
Let's talk
Let's work out the time your ward can give back to care
15 minutes to identify the most valuable use — hosted in France, the HDS requirement settled at scoping, supervised, with no commitment.