Appointments, visit prep and routine requests handled 24/7, on your EHR, with staff handoff built in.
We build AI virtual health assistants that book and reschedule appointments, collect intake and insurance before the visit, take refill requests for staff review and answer questions from approved content, by chat, SMS, portal or voice, with identity checks before any PHI is shown.
Workflow scoping · Evaluation design · Human oversight · Integration planning
Updated September 2026
An AI virtual health assistant takes the routine requests that fill phone queues and completes them through your systems: it finds an open slot and books the visit, sends preparation instructions, checks insurance, logs a refill request for staff, and answers questions from content your team approves. Anything clinical goes to triage or the care team with the conversation attached, so patients get a fast answer and staff keep every decision that needs judgment.
A patient-facing assistant earns trust by finishing routine tasks and knowing exactly where its job ends. These are the tasks we build it to complete end to end.
Searches open slots in the EHR by visit type, provider, location and time window, books or holds one, confirms by SMS or email, and releases the slot on cancellation so it can be filled from a waitlist.
Collects demographics, insurance card images and intake questionnaires before the visit, checks eligibility through an X12 270/271 inquiry, and tells the patient what to bring and when to arrive.
Answers questions about locations, hours, services, preparation instructions, billing and coverage from content your team approves, cites the source page, and captures unanswered questions so content gaps are visible.
Takes a refill request, confirms the medication against the active medication list, and creates a task for clinical staff. The assistant never approves a refill or changes a prescription.
Sends appointment, preparation and follow-up reminders on the channel each patient consented to, handles replies such as confirm or reschedule, and stops on opt-out.
When a patient describes symptoms, the assistant hands the conversation to your triage workflow or nurse line with the transcript attached instead of giving medical advice.
One conversation engine serves every channel. Voice runs over your telephony platform with speech recognition and barge-in, and every channel offers a transfer to a person.
General questions need no login. Anything patient-specific requires a verified identity: portal single sign-on, or a match on name and date of birth plus a one-time code to the phone or email on file. Proxy access for parents and caregivers follows the permissions in the EHR.
SMS reminders run on registered 10DLC numbers with recorded consent and STOP handling, in line with TCPA rules, and message content follows the minimum necessary standard.
The assistant converses in the languages your patients use, offers interpreter handoff, and the web widget meets WCAG 2.1 AA, which supports your language access and accessibility duties under Section 1557.
The assistant reads and writes through the EHR's APIs so staff see every booking, request and message in the systems they already use.
Patient for identity matching, Slot and Schedule for availability, Appointment for booking, Coverage for insurance, MedicationRequest for refill context, QuestionnaireResponse for intake, and Communication or Task for messages that staff must act on.
Epic's FHIR R4 scheduling and patient APIs with SMART on FHIR authorization, and MyChart-linked flows where the organization enables them.
athenahealth's appointment and patient APIs alongside its FHIR endpoints, and eClinicalWorks FHIR reads with HL7 v2 SIU scheduling messages where write access calls for it.
Oracle Health Millennium scheduling APIs, and practice management systems through Interoply connectors, our integration platform for FHIR, HL7 v2 and X12.
Answers come only from approved content. Questions about symptoms, test results or treatment route to triage or the care team, and the assistant says so plainly.
The assistant can call only the tools it needs: slot search, booking, eligibility, refill task creation and message creation. It cannot read notes, results or anything outside its scope.
Every release runs a red-team suite that tries to extract another patient's data, bypass identity checks or override instructions. A single failure blocks the release.
Generative AI messages about clinical information carry the disclosure and human contact path California AB 3030 requires, and the configuration supports other state disclosure rules such as Texas HB 149.
Contact center and portal vendors offer assistants for common scheduling and FAQ flows inside their own platforms. They fit when your workflows match their templates and one EHR is in scope.
Build when the assistant is part of your own product, spans several EHRs or practice management systems, needs specialty scheduling rules such as referral-required visits, or must follow your own approval and escalation model.
Task flows, handoff rules, approved content set and the out-of-scope policy, signed off by operations and clinical leads.
Channel adapters, identity layer, tool gateway, EHR connectors and the staff task queue, in your cloud or ours.
Task scripts, identity and disclosure red-team cases, and release gates wired into CI.
Dashboards for completion, handoffs and unanswered questions, plus a runbook and content update process.
Map the top request types from call and message logs, confirm EHR access, identity method and handoff owners.
Scheduling, intake and refill flows against a non-production EHR, with the evaluation suite running on every change.
Release on web chat or SMS for one service line, with daily review of handoffs and unanswered questions.
Add voice, more visit types and more locations once the pilot measures hold.
Visit types, provider rules, referral requirements and multi-location templates drive most of the build effort.
Each EHR or practice management system adds its own read and write paths.
Voice adds telephony, speech and latency work; SMS adds 10DLC registration and consent handling.
The size of the approved content set and the number of languages to evaluate.
A patient app that answers by voice or chat only from the patient's own record and a cited clinical guideline, and never poses as the care team. Client name withheld.
We sign BAAs, and Nirmitee is ISO 27001:2022 certified.
How scheduling agents, intake agents and patient-facing AI are built and evaluated.
Scheduling automationAI agents vs chatbotsAll healthcare AI agents
An AI virtual health assistant is a patient-facing agent that handles routine requests by chat, SMS, portal or voice: booking and rescheduling appointments, visit preparation, answers from approved content, refill requests and reminders. It works through the EHR's APIs and hands anything clinical to your staff.
Yes. It searches open slots through the EHR scheduling API by visit type, provider, location and time window, books or holds the slot, and sends a confirmation. Visit types that need a referral or a staff decision go to a scheduler queue instead of being booked automatically.
Yes. Scheduling, rescheduling, visit preparation, refill requests and content questions run around the clock. Requests that need a person outside business hours go to the on-call workflow for anything urgent and to a staff queue with a callback time for everything else.
No. It answers only from content your team approves. When a patient describes symptoms or asks about results or treatment, it hands the conversation to your triage workflow or care team with the transcript attached, and tells the patient who will respond and when.
General questions need no login. Patient-specific tasks require portal single sign-on or a match on name and date of birth plus a one-time code sent to the phone or email on file. Proxy access for parents and caregivers follows the permissions already set in the EHR.
Epic, Oracle Health, athenahealth and eClinicalWorks through their FHIR R4 APIs, plus athenahealth's own APIs and HL7 v2 scheduling messages where write access calls for them. Practice management systems connect through Interoply, our integration platform.
It is when the whole system is built for PHI: BAAs with every party including the model provider, identity checks before PHI is shown, encryption in transit and at rest, minimum necessary tool access, and audit logs. We sign BAAs and Nirmitee is ISO 27001:2022 certified.
Yes. The same conversation engine runs over your telephony platform with speech recognition, interruption handling and transfer to a person at any point. Voice usually follows a chat or SMS pilot so flows are proven before latency and speech are added.
We track task completion by request type, correct handoffs, unanswered questions, abandonment by step, time to staff response for handed-off requests, and zero tolerance for unauthorized disclosure in red-team tests. Conversation volume alone is not a success measure.
Some do. California AB 3030 requires a disclaimer and a way to reach a person on generative AI communications about clinical information unless a licensed provider reviews them, and Texas HB 149 requires disclosure of AI use in care. We build disclosure text and routing as per-state configuration.
About 9 to 12 weeks to a live pilot: 2 weeks of discovery, 4 to 6 weeks of build in the EHR sandbox, and 3 to 4 weeks of pilot on one channel and service line. EHR app review and SMS number registration run in parallel.
More Agents
Explore complementary solutions for your practice.
Symptom intake, red-flag screening and acuity routing, built on your protocols and your EHR.
Worklists, device triage, outreach and billing evidence for CCM and RPM programs, with clinicians setting every rule.
Bring your top request types, your EHR and your handoff model. We will map the flows, identity method, integrations and pilot plan for your first channel.
