AI voice agents for telehealth book virtual visits, call patients who have not confirmed, walk them through a quick tech check and hand off to the follow-up calls after the visit. The join link itself goes out from the clinic's own system. The agent verifies the patient before naming the visit and sends every clinical question to a person.
Key Takeaways
- The join link goes out by text. The call is for patients it missed.
- Verify the patient before naming the visit or the clinician.
- Most virtual visits are behavioral health, so build the crisis route first.
This post is part of our guide to AI voice agents for hospitals and health systems. The guide sorts patient calls by care setting first, and telehealth is the setting where the visit happens on a screen while the calls around it still happen by phone. This post builds those calls in Dograh, an open-source, self-hostable voice agent platform: the booking, the reminder call for patients who have not confirmed, the tech check and the hand-off after the visit.
Most virtual visits are behavioral health, so plan for that first
Telehealth is a steady part of care, and many of the patients on the line are anxious. FAIR Health tracks telehealth use across its national claims data. In its telehealth trends for January to March 2025, 14.9% of patients had a telehealth claim in January. Of those patients, 58.5% had a diagnosis of a mental health condition, and the share rose to 62.3% in February.
That second number shapes the whole design. A telehealth line is mostly a behavioral health line, so even a simple reminder call can reach a patient who is having a hard day. The agent needs a crisis route before it needs a clever booking flow. The post on AI voice agents for behavioral health intake and triage covers that route in depth, and the rules in this post borrow from it.
Book the visit, then send the link from your own system
Booking a virtual visit runs the same way as booking an office visit, against the live calendar. The agent verifies the patient, reads open slots from the scheduling system through an HTTP API tool, books the slot the patient picks and writes it back. The mechanics, including reschedules and cancellations, are in AI voice agents for patient appointment scheduling.
The join link is where telehealth differs. Dograh has no SMS tool and sends no texts itself. The link should go out from the clinic's own telehealth or messaging system the moment the visit is booked, with no voice agent involved. If a patient asks on a call for the link to be sent again, the agent can call your messaging provider's endpoint through an HTTP API tool, and your provider sends the text.
Keep the link out of the conversation itself. A spoken web address is hard to catch and easy to mistype, and a join link often carries a code tied to that one visit.
The call is for the patient the reminder did not reach
Routine reminders work well as texts and emails, and a call earns its place only when those have not worked. Your telehealth system already knows which patients confirmed the visit and which finished the device test. The patients who did neither are the ones to call. Your scheduling system starts each call through Dograh's API Trigger a day or a few hours before the visit, passing the visit time and a patient reference into the call.
On that call the agent runs a short tech check. It asks whether the link arrived and whether it opens on the phone or laptop the patient plans to use, including the browser's request for camera and microphone access. If the patient cannot get it working, the agent offers a staff callback or a new time. It offers an audio-only visit only when your system returns that as an option, because running a visit without video is a clinical and billing decision for the clinic, never the agent's.
An appointment reminder to a patient's mobile generally falls under the Federal Communications Commission (FCC) healthcare exemption, with strict limits on how often you call, and the guide's consent section quotes the rule. Reminder calls that cut missed appointments more broadly are covered in Healthcare No-Show Reduction: Why AI Voice Agents Beat Reminders.
Open Source Alternative to Vapi / Retell
Self-hosted voice agent platform — no per-minute fees
dograh-hq/dograh
Star on GitHub
Greet, verify, then talk about the visit
Every call runs greet, verify, disclose, and the order matters more on a telehealth line. In Dograh, the Greeting Text plays word for word before verification. Write it to name the clinic and say the patient is speaking with an automated assistant; Dograh plays exactly what you write. It can carry variables with a fallback, such as {{clinic_name | your care team}}, but never the visit type or the clinician's name, since a therapist's name alone can reveal why someone is being seen.
Verification comes next: date of birth plus one reference, such as the appointment date or part of the medical record number. A phone number finds the record. It does not prove who answered. The Health Insurance Portability and Accountability Act (HIPAA) Privacy Rule at 45 CFR 164.514(h) requires a provider to verify the identity of a person requesting protected health information (PHI) when that person is not known to it. Only then does the agent name the visit.
From there the agent collects, reminds and routes, and it never gives medical advice. A symptom or a clinical question goes to a nurse or clinician through Dograh's Transfer Call, which works on Twilio only and passes no summary, so the agent writes the details to your system before it transfers. If nobody picks up within the timeout, 30 seconds by default, the patient stays with the agent, which offers a callback and writes the request through an HTTP API tool.
An emergency gets an instruction to hang up and call 911, and crisis language gets the 988 Suicide and Crisis Lifeline. Both are fixed wording once the emergency pathway is chosen: a pathway's Transition Speech plays fixed text or a recorded clip exactly as written. Choosing the pathway is still the language model reading a written condition, so keep that phrase list short and explicit, and test it on real calls. Text Transition Speech does not play on speech-to-speech models, so use a standard agent or a recorded clip for these lines.
After the visit, hand off to two separate calls
The visit ends, and two different calls can follow it. The first is a clinical check-in that asks how the patient is doing and whether the next steps are clear, with any symptom routed to a nurse. That is AI voice agents for post-visit follow-up in telehealth. The second is the experience survey, one rating and one open question, covered in AI voice agents for patient surveys in telehealth.
Keep them apart. A patient who has just described a worrying symptom should not be asked to rate the visit out of five in the same breath, and a rating mixed into a clinical record muddies both. Each call starts from the visit's closing event in your telehealth platform or electronic health record (EHR) through the API Trigger, and each writes its answers back by webhook.
Join the Dograh Community
Dograh is an OSS alternative to Vapi. Join our Slack community for queries, releases, best practices & community interactions.
Keep virtual-visit data on your own servers
A telehealth call carries PHI from its first sentence, and in a caseload that is mostly behavioral health, that data is among the most sensitive a provider holds. Self-hosting Dograh runs the platform on your own servers, so recordings and transcripts sit next to your EHR. No business associate agreement (BAA) with Dograh is needed in that setup, because Dograh never receives the PHI. Model providers you use through an API key still hear the audio, so each needs its own BAA. The phone carrier may need one too, depending on what it stores. A BAA delegates duties to the vendor and never transfers your liability. Patient audio stays inside your infrastructure only when the speech and language models also run on your servers.
There is no retention setting and no automatic deletion, so your own policy decides how long recordings and transcripts are kept. On a self-hosted server, recording can be switched off for the whole server, and transcripts are still saved. Self-hosted Dograh has no per-minute platform fee, though servers, any paid model keys and phone minutes still cost money. Dograh Cloud has a 1¢-a-minute platform fee.
Start with the reminder call for unconfirmed visits. It reads from one system and writes very little, and its transcripts will soon show whether patients are struggling with the link or with something else.
Glossary
- API Trigger
- Dograh's way for another system, such as a scheduling or telehealth platform, to start an outbound call and pass details like the visit time and a patient reference into it.
- audio-only visit
- A telehealth visit held by phone with no video. Whether one is offered is the clinic's clinical and billing decision, and the agent offers it only when the clinic's system returns it.
- Greeting Text
- The optional fixed opening line of a Dograh call, spoken word for word before verification. It can name the clinic but must never carry an appointment or a clinician's name.
- Transition Speech
- Fixed text or a recorded clip that Dograh plays exactly as written when the agent moves onto a pathway, which makes it the right place for an emergency or crisis instruction.
