Visit calls that verify before saying anything
Dograh is the self-hostable voice agent platform with an open-source workflow builder. Run the calls around every virtual visit on your own servers, with the agent naming only the practice until the patient passes your check.
Greet, verify, then disclose
The opening names the practice and nothing else
Greeting Textplays word for word before any check
A clinician's name can reveal as much as a diagnosis, so the first line leaves it out.
→ Switch off Allow Greeting Interruption if it carries a notice
Your endpoint decides who is verified
HTTP API toolasks your system during the call
The agent collects date of birth plus one more item on file, and your endpoint replies pass or fail. Only a pass unlocks the visit time.
→ Build and test this endpoint first
Device checks from the patient portal
Voice widgeta live voice call inside your web page
A patient who starts the call from your portal is already testing that browser's microphone, with no phone carrier in the path.
→ Keep the visit type out of tech checks
What the agent reads, writes and runs on
Pass a patient reference at call start, never the visit type
Tools send a visit reference, so the link never reaches a log file
The nurse answers before the patient joins
Check these before a patient hears the agent
- 911 line on an emergency pathwayPut the fixed line in Transition Speech so it plays word for word, then test it with many phrasings.
- Telemetry switched offSet ENABLE_TELEMETRY=false and remove POSTHOG_API_KEY before the first visit call.
- Every hosted model provider on your BAA listHosted speech and language models receive the audio. Open-weight models on your servers keep it in-house.
- Reminder calls kept inside the exemptionEach lasts about a minute, a patient gets at most one a day and three a week, and balances never come up.
- Retention schedule set on storageRecordings, transcripts and logs stay only as long as your own schedule says.
Patient audio where you control access
Every copy in storage you run
Self-hosted, recordings, transcripts and call records land in your own object storage and Postgres.
No BAA with Dograh needed
The platform never receives your call data when it runs on your servers.
Open-weight models keep audio home
Run speech and language models on your own hardware, and no model vendor hears the visit call.
Nurse line reached first
Transfer Call dials the nurse first, and a Twilio introduction can stay at the patient's first name and topic.
What changes when the platform sits with the provider
| What you get | Dograh | hosted per-minute voice platforms |
|---|---|---|
| Self-hosted inside your own environment | ✓ | ✕ |
| No per-minute platform fee on your own servers | ✓ | ✕ |
| Open-source code your compliance team can review | ✓ | ✕ |
Healthcare privacy, call by call
Does the agent read the join link aloud?+×
No. Your telehealth or messaging system sends it. If a patient asks again, the agent asks your messaging provider to resend it.
Can a tech-check call expose health information?+×
It can if the visit type is read aloud, since that reveals why the patient is seen. The agent sticks to the device and browser.
Do reminder calls need written consent?+×
Reminder calls to mobiles can generally rest on the TCPA healthcare exemption, within its limits. Confirm your setup with counsel.
What happens if a patient describes an emergency?+×
A fixed line on an emergency pathway tells them to call 911 now. Test that pathway with many phrasings before go-live.
Do I need a BAA with Dograh?+×
Not when you self-host, because the platform never receives your call data. Check your carrier, cloud and hosted model providers.
Does a web call skip the phone carrier?+×
Yes. The voice widget runs over WebRTC, so no carrier is in the path, though hosted models still receive the audio.
Build the verify endpoint first.
Self-host in one command, or start on Dograh Cloud.
$ curl … start_docker.sh && ./start_docker.sh