CuraOS — the agentic healthcare operating system.
An AI-native, multi-tenant EHR for clinics, hospitals and diagnostic centres, in production for two years. Agents draft the notes, summaries, prescriptions and follow-ups; a named clinician reviews and signs. Built on HL7 FHIR R4 and designed for ABDM’s consent-based exchange.
CuraOS in two minutes.
How a consultation becomes a structured note, how the record follows the patient, and where the clinician signs off.
Interface shown is illustrative. No sign-up required to watch, and the video is served from our own servers — nothing is sent to a video platform.
An EHR with agents inside it — and a clinician on every consequential decision.
Most hospital software stores what a clinician typed. CuraOS works the other way round: the record is the spine, and specialised agents do the drafting, checking, coding and chasing that otherwise eats a doctor’s evening. It has been in production for two years — in clinics, in diagnostics centres, in remote patient monitoring, and in longevity and preventive-health programmes.
One record per patient
Duplicate detection at registration, so a returning patient is matched rather than recreated. ABHA linking where the patient consents. Visits, admissions, labs, imaging, prescriptions and bills on one timeline instead of in four systems that disagree.
Agents that draft, not decide
The scribe writes the SOAP note. The prescription assistant checks allergies, interactions and duplicate therapy. The lab agent marks what is out of range against the patient’s own prior values. Each output arrives as a draft with its evidence attached.
Traceable and accountable
Every AI output records the inputs it read, the model version that produced it, and the clinician who accepted, edited or rejected it. That trail is part of the record, because the clinician who signs is accountable for the decision.
What we will not do is dress that up. We publish no customer names, patient volumes, uptime figures or time-saved percentages, and we hold no regulatory certification. On a demo call we show the working system, say what is scoped next, and if what you need is outside our scope we will tell you on the call.
See the working systemThe record system is rarely the problem. The documentation load is.
A doctor who has seen forty patients has produced forty consultations worth of clinical reasoning and perhaps eight notes worth of typing. The gap is arithmetic, not laziness. So summaries get written late, from memory; allergy checks depend on the prescriber remembering; follow-ups depend on someone having time to call.
CuraOS carries that load — and holds the boundary. Drafting documentation is something AI is genuinely good at. Deciding what is wrong with a patient is not a problem we hand to a model.
See where the human checkpoint sitsSound familiar?
- Doctors finish the list and then spend an hour typing notes the consultation already contained.
- Discharge summaries are assembled two days later by whoever is on duty, from a file and a memory.
- Lab, imaging and billing each hold a slightly different version of the same patient.
- A patient arrives with a prescription from another hospital and nobody can see what was already tried.
- Coding queries come back weeks later, when the clinician has forgotten the case.
- ABDM and ABHA came up in a tender and nobody can say what your current system supports.
- Records sit with an HIS vendor whose contract says nothing about deletion, access logging or breach notification.
Nine AI modules, working today.
These are the modules in production, not a feature list waiting on a release. We publish the boundary between what runs today and what is next, because it is the first question a hospital IT head asks and you should hear it from us rather than find it in month three.
Patient registration & ABHA linking
Duplicate detection across name, phone, date of birth and identifiers, so one patient stays one record. ABHA linking where the patient holds a number and consents to the link.
Unified longitudinal record
One patient, one timeline: encounters, diagnoses, medications, allergies, investigations, imaging, procedures and documents, each attributed to whoever created it.
AI medical scribe
Ambient capture of the consultation, turned into a structured SOAP note. The doctor edits and signs; an unsigned draft is not a clinical record.
AI clinical summary
A cited read of a long record for the clinician who has ninety seconds: active problems, current medications, recent results and what changed, each line linked to its source.
AI prescription assistant
Drafts from the recorded assessment and checks allergies, current medications, duplicate therapy and dose concerns, showing the basis for each flag. The prescriber confirms or overrides every line, and overrides are recorded.
AI lab report analyser
Reads structured results and uploaded reports, marks values outside reference range, compares against the patient’s own history and raises trends worth attention. Interpretation stays with the clinician.
AI discharge summary
Assembles the summary from admission and progress notes, procedures, medications and investigations already in the record, in your format. The treating physician signs before it is issued.
AI patient chatbot
Handles what fills a front desk: appointment slots, preparation instructions, whether a report is ready. Scoped to non-clinical queries by design, and routes anything clinical to the care team.
AI follow-up agent
Tracks who was asked to return and when, reminds them on consented channels, and flags the patients who did not come back — the group a busy clinic loses first.
Second phase and advanced expansion.
The roadmap of a product already running. Everything below is decision support with mandatory clinician review: none of it issues a diagnosis, a prescription or a claim on its own, and that does not change as the scope grows.
Second phase
- Diagnosis support and clinical decision support
- Radiology assistant for report drafting and prior-study comparison
- Medication reconciliation across admission, transfer and discharge
- Care plan generator and chronic disease management
- Referral agent, with the clinical context attached
- ICD and SNOMED coding assistant — for the coder, not instead of them
- Billing assistant reconciling orders, tariffs and documentation
Advanced expansion
- Insurance and pre-authorisation drafting — the agent assembles the pack, the medical officer submits it
- Population health across a defined cohort, for the people who plan capacity
We do not quote dates for modules that have not entered build. What we will give you is the current sequence and what moves it — usually a deployment’s integration requirements rather than our engineering queue.
Each agent has a job, a boundary and a named human who signs.
The diagnostic agent assembles considerations and the findings behind them without independently issuing diagnoses. The medication agent raises flags and the prescriber clears them. The coordinator talks to patients about logistics, not about their condition. This is the discipline the product is built around, and the part we will not trade for a better demo.
Doctor Copilot
Pulls the relevant history forward, drafts the note and the orders, and answers questions about this patient’s record with links back to the entries it used.
Human checkpoint: the doctor signs every note and order before it becomes part of the record.
Ambient Clinical Scribe
Listens in the room and turns a consultation in Hindi, English or another supported language into a structured note, including medications and investigations discussed.
Human checkpoint: the clinician sees transcript and draft, edits, and signs. Recording requires the patient’s consent, captured first.
Diagnostic Intelligence
Assembles the considerations the record supports, shows the findings behind each, and points at what has not been ruled out — without independently issuing diagnoses.
Human checkpoint: the diagnosis is the clinician’s. The agent has no path to enter one into the record.
Medication Safety
Checks allergies, interactions, duplicate therapy and dose concerns including renal and paediatric considerations, and reconciles medication lists across admission, transfer and discharge.
Human checkpoint: the prescriber confirms or overrides each flag, and an override is recorded with its reason.
Patient Care Coordinator
Appointments, preparation instructions, report availability, follow-up reminders and chronic-care check-ins, in the patient’s language and on the channel they agreed to.
Human checkpoint: it gives no clinical advice. Clinical questions route to the care team with the context attached.
Clinical Quality & Compliance
Finds incomplete documentation, missing consent records, coding gaps and deviations from the hospital’s own protocols, and feeds them into quality review.
Human checkpoint: findings are raised as queries to the quality team. No agent edits a clinical record.
The hospital still has to run. CuraOS covers the departments, not only the AI.
An AI layer on top of software nobody uses is shelfware. These are the screens a nurse, a lab technician, a pharmacist and a billing clerk live in all day. Which modules a deployment switches on is decided at scoping.
OPD
Appointments, queue, consultation, orders and prescriptions in one flow.
IPD
Admission, bed and ward management, progress notes, transfers and discharge.
Emergency
Triage, rapid registration for unidentified patients, timed event capture.
Nursing
Vitals, medication administration, handover and tasks, on a console built for a shift.
Laboratory (LIS)
Order to sample to result to verification, with analyser interfacing where available.
Radiology (RIS / PACS)
Scheduling, reporting and DICOM study access beside the clinical record.
Pharmacy
Dispensing against the prescription, stock, batch and expiry, and indents.
Surgery & OT
OT scheduling, pre-operative checklists, operative notes, implant records.
Billing
Tariffs and packages, estimates, interim bills and settlement tied to clinical orders.
Insurance & TPA
Pre-authorisation and claim packs assembled from the record, on logged channels.
Telemedicine
Consultations with the same record, notes and prescription flow as an OPD visit.
Patient portal
Reports, prescriptions, appointments, bills and the patient’s own consent history.
Hospital analytics
Occupancy, throughput, revenue, turnaround times, documentation completeness.
Research
De-identified cohort extraction for approved studies, with the approval recorded.
The capabilities that make CuraOS different — with their status stated.
This is the part of a product page where most vendors blur the line between shipped and imagined. The table below does not. Status means what it says: live today, scoped for a later phase, or dependent on validation that does not yet exist.
| Capability | What it does | Status |
|---|---|---|
| Ask Your EHR | Natural-language search across the records a user is authorised to see, with every query logged. | Live today |
| Clinical knowledge assistant | Answers clinical reference questions with citations, so the clinician can check the basis rather than trust a paraphrase. | Live today |
| Voice-first EHR | Navigation, order entry and note dictation by voice, for clinicians whose hands are occupied. | Second phase |
| AI prior authorisation | Assembles the pre-authorisation pack against the insurer’s requirements. The medical officer reviews and submits. | Second phase |
| Predictive deterioration alerts | Flags admitted patients whose trajectory is worsening, ahead of the next round. Needs local calibration and clinical review of thresholds first. | Roadmap · validation required |
| Multimodal medical AI | Reading images, documents, waveforms and text together rather than as separate silos. | Roadmap |
| Clinical trial matching | Identifies patients who may meet a trial’s criteria. Nothing reaches a sponsor without the patient’s specific consent. | Roadmap · consent-gated |
| Population health intelligence | Cohort-level view of prevalence, screening coverage and care gaps. | Advanced expansion |
| Patient digital twin | A model of one patient used to reason about likely response to a change in therapy. A research direction, and we say so: it informs no care decision until clinical validation exists. | Roadmap · subject to clinical validation |
| Federated analytics | Learning across sites in a group without centralising identifiable patient data — what makes multi-site analytics defensible under the DPDP Act. | Roadmap |
Why the system is shaped this way.
Two decisions drive most of the design: one tenant’s data must be unreachable from another, and every read and write — by a human or an agent — must be reconstructable afterwards. In healthcare those are not features. They are the conditions under which everything else is allowed to exist.
Interfaces built per role
A web EHR for the desk, a doctor mobile app for rounds and on-call, a patient portal, and a nursing console designed around a shift rather than a form. One record underneath, four ways in.
One gateway, no side doors
Every request — app, portal, agent or integration — passes an API gateway handling OIDC authentication, role- and attribute-based authorisation, tenant scoping, consent checks and audit writing. Clinical services do not implement their own access rules; that is how inconsistencies become breaches.
Tenant isolation as a boundary, not a column
Multi-tenancy is where shared healthcare platforms fail most expensively. Tenant identity is enforced at the gateway and carried into every query, with isolation at the data layer rather than a filter an engineer can forget.
An audit layer that answers questions
An append-only log records who or what accessed which record, under which consent, and what changed. It exists to answer the three questions that actually get asked: who saw this file, which model version produced this draft, and which clinician signed it.
Agent orchestration, separate from the record
Agents run in an orchestrator beside the clinical services, not inside them. They read only what the requesting user may read, write drafts rather than records, and each run is versioned and logged — so an agent can be paused or rolled back without touching clinical data.
Interoperability as a layer of its own
A FHIR layer handles ABDM and ABHA flows, HL7 FHIR R4 resources, LIS, RIS and PACS interfaces, and external EHR APIs. Keeping translation out of the clinical core is what lets an integration change without a clinical release.
A data layer chosen for the job
PostgreSQL with pgvector for the record and for retrieval over authorised documents, object storage for images and files, an event queue for asynchronous work, and the audit log as a first-class store.
Tested the way we test clients’ systems
Avisha Global’s other business is AI security. The people who run prompt-injection and agent-permission testing for enterprise clients review CuraOS — an agent with access to patient records is exactly the system we are usually hired to break.
One department first. Expand on evidence.
Nobody should move a hospital onto a new record system in one weekend. A deployment runs in four stages with a review gate between each, and the longest stage is almost always integration rather than software.
Scoping
Departments, specialities and existing systems mapped. Data-protection terms, hosting location and the integration list agreed in writing before configuration starts.
Configuration & integration
Tenancy, roles, templates and tariffs. LIS, RIS and PACS interfaces built and tested. ABDM integration, where in scope, goes through sandbox validation and security review — a process, with its own timeline.
Pilot department
One OPD or one ward goes live with clinician training and on-site support. Agent drafts are reviewed against clinician edits, so you see how much editing a draft needs.
Rollout & review
Remaining departments in waves, with historical record migration as agreed, then a review against the measures set at scoping — your measures, not a published average.
A single doctor and a hospital group need different things from the same core.
CuraOS is one multi-tenant platform, configured per deployment. A two-doctor clinic should not have to think about bed management, and a 400-bed hospital should not live inside a clinic app. What changes is which modules are switched on, which integrations exist, and how the roles are drawn.
Two years of production use has stretched that core where conventional hospital-management software does not go: remote patient monitoring, and longevity and preventive health, where the record keeps accumulating between visits rather than only during them.
See how it is packaged
Small clinics in India
In production today. Registration, consultation, scribe, prescriptions with safety checks, billing and ABHA linking — so the doctor leaves with the notes already written.
Multi-specialty clinics
Several practitioners on one record, referrals between departments with context attached, in-house lab and imaging orders, and role separation that holds.
Mid-sized hospitals
IPD and wards, nursing console, OT, pharmacy, TPA workflows, discharge summaries ready at discharge, and documentation-completeness reporting for quality review.
Hospital chains
Per-unit tenancy with group reporting, shared masters and tariffs, and analytics across units without pooling identifiable patient data where that is not permitted.
International providers
The same core on HL7 FHIR R4 with ABDM-specific flows off, multi-currency billing, configurable terminology and formats. Local privacy requirements assessed per jurisdiction first.
Diagnostics centres
In production today. Order to sample to verified report, analyser and DICOM interfacing, the lab report analyser, and consented report delivery.
Remote patient monitoring
In production today. Readings arrive between visits onto the same record, and thresholds raise a query to a named clinician rather than acting on their own.
Longevity & preventive health
In production today. Repeat panels, risk tracking and care plans over years rather than episodes — the case that breaks software built around a billable admission.
Home healthcare
Visit scheduling, mobile capture of vitals and notes, care plans, the follow-up agent, and a portal for families who need to see what happened today.
Three packages. Pricing scoped per deployment.
What drives cost in healthcare software is user count, module scope, integrations and migration — not a number we can put on a web page. So we do not publish one. After a scoping call you get a fixed, written quote for a defined scope.
Starter
For independent doctors and small clinics.
- Registration, longitudinal record and OPD consultation
- AI medical scribe and clinical summary
- Prescription assistant with allergy and interaction checks
- Billing, patient portal and the follow-up agent
- ABHA linking and the patient chatbot
Professional
For multi-specialty clinics and diagnostic centres.
- Starter, across multiple practitioners and departments
- Laboratory (LIS) and radiology (RIS, PACS, DICOM)
- Lab report analyser and the referral agent
- Telemedicine and referring-doctor access
- Ask Your EHR over authorised records, plus clinic analytics
- Second-phase decision-support modules as they release
Enterprise
For hospitals and hospital groups.
- Professional, plus IPD, emergency, nursing, OT and pharmacy
- Insurance and TPA workflows, and discharge summary generation
- Per-unit tenancy with group reporting and shared masters
- Integration programme for existing LIS, RIS, PACS and ERP systems
- ABDM integration, subject to sandbox validation and security review
- Clinical quality and compliance agent, and research extracts
- Named implementation team, data migration and go-live support
Bring your department list, your existing systems and the integrations you cannot do without. We will tell you which package fits, what the integration work really involves, and what is not available yet — before anyone talks about commercials.
Scope a CuraOS deployment
Patient data is the most sensitive category this company handles. We already do this work for hospitals.
Avisha Global runs DPDP compliance programmes for hospitals, insurers and health platforms, so we have seen what goes wrong: blanket admission consent covering "any other purpose", HIS vendors with unlogged access to production, discharge summaries sent to TPAs over WhatsApp, records kept forever because nobody wrote a schedule. CuraOS is built by the people who write those findings — and consent has to follow the record once it leaves you for a lab, an insurer or a TPA.
- Consent recorded as versioned data, carried with the record when it moves
- Purpose-scoped access — treatment, billing, research and marketing are distinct bases
- Retention per category, separating clinical records held under medical record-keeping rules from operational and marketing data
- Rights requests — access, correction, erasure — as a workflow with an owner, not a support ticket
- Guardian consent for minors, and defined treatment of attendant and emergency-contact data
- Breach playbook covering vendor incidents and patient notification, not only IT recovery
- Hosting location agreed in writing, with processor terms on security, sub-processing, deletion and breach duties
Frequently asked questions
See CuraOS on your own departments.
A 30-minute demo with a product specialist, using your speciality mix and your existing systems — on the live product, not a slide deck. We will also tell you what is next and what we cannot do.