Zoom RTMS & Live Transcription

Healthie's AI Scribe uses Zoom's Real-Time Media Streams (RTMS) to capture live session transcripts during telehealth appointments. RTMS replaced the legacy cloud recording approach and is the current standard for all new Scribe enablements.

Rather than polling the Zoom API after a session ends, RTMS is fully push-based: Zoom sends a single webhook when a session starts, Healthie's Media Streams service connects to the stream directly, and the transcript is written incrementally to storage as the session progresses.

AI Scribe & Transcript Integration

Healthie's AI Scribe automatically captures and transcribes telehealth sessions, turning real-time Zoom conversations into structured, searchable transcripts, without any manual recording or transcription effort from providers.

Live, real-time transcription during Zoom sessions
Automatic delivery of full transcripts post-session
Webhooks to trigger downstream workflows (note generation, charting, summarization)
Scribe controls scoped per appointment type — patients are only notified when transcription is actually active

Transcripts Only

If you're interested in transcript access without the full AI Scribe feature (e.g., to power your own downstream AI, charting tools, or workflows), reach out to hello@gethealthie.com to learn more.

RTMS vs. Cloud Recording

	RTMS (current)	Cloud Recording (legacy)
Trigger	Push-based (one webhook per session start)	Polling the Zoom API for recording URLs
Rate limits	None — no per-recording API calls	Subject to 429 errors under high volume
Transcript delivery	Incremental + post-session webhook	Post-session only (can take hours)
Client notification	"Content of this meeting is being shared with app(s)" modal	"This meeting is being recorded" banner
Transcript quality	Identical — both use Zoom's transcription engine	Identical

Note: RTMS receives transcript text directly from Zoom — Healthie does not run separate audio transcription. The word-level output is identical to cloud recording transcripts.

Transcript Pipeline

During the session

Provider starts a Zoom session with AI Scribe enabled
Zoom initiates the RTMS stream and fires appointment.transcript_started
Healthie's Media Streams service connects and begins receiving transcript data
Transcript is written incrementally to S3 as the session progresses — readable mid-session once appointment.transcript_started has fired

After the session

Session ends; Zoom completes transcript processing
appointment.transcript_available fires — typically within ~2 minutes of session end
Full transcript is available in S3 in VTT format, fetchable via the Healthie GraphQL API

Recommendation: For most use cases, wait for appointment.transcript_available before fetching. Mid-session access is supported but the stream may be incomplete.

Webhooks

Healthie sends the following webhook events for AI Scribe / RTMS sessions. These are separate from the legacy cloud recording events.

RTMS / Real-Time Transcript Events

Event	When it fires	Use for
appointment.transcript_started	RTMS stream goes live (session in progress)	Confirming stream is active; mid-session transcript access
appointment.transcript_available	Full transcript written to S3 (~2 min post-session)	Triggering transcript processing workflows
appointment.transcript_failed	Stream fails for any reason	Alerting, retry logic; includes a failure reason

Cloud Recording Events (legacy — still available)

Event	When it fires
appointment.recording_started	Recording begins in Zoom
appointment.recording_stopped	Recording ends in Zoom
appointment.transcript_available	Cloud transcript ready (can take hours post-session)

Participant Events

Event	When it fires
appointment.participant_joined	Someone joins the Zoom session (includes name + email)
appointment.participant_left	Someone leaves the Zoom session

Note: appointment.transcript_available is shared between RTMS and cloud recording paths. It fires for whichever transcript source is active on the account.

Webhook Payload Structure

Healthie uses thin payloads — webhooks contain minimal identifying data. After receiving a webhook, fetch the full resource via the Healthie GraphQL API.

Key fields:

resource_id — The appointment ID
resource_id_type — Always "Appointment" for Scribe events
event_type — The specific event that occurred

Setting up Webhooks

Register via GraphQL

Basic Endpoint (Node.js)

Fetching Transcript Content

After receiving appointment.transcript_available, query the Healthie GraphQL API for the appointment details and transcript.

Transcripts are stored in VTT (WebVTT) format and accessible from the appointment record. They can be opened in any standard text editor.

Security

Webhook Signature Verification

Healthie signs all webhooks using HTTP Message Signatures (RFC 9421). Always verify signatures before processing.

Relevant headers:

Content-Digest — SHA-256 hash of the payload
Signature-Input — Specifies signature components
Signature — The cryptographic signature

IP Whitelisting

Configure your firewall to accept webhooks only from Healthie's IPs:

Environment	IP Addresses
Production	52.4.158.130, 3.216.152.234, 54.243.233.84, 50.19.211.21
Staging	18.206.70.225, 44.195.8.253

Key Constraints & Notes

Transcript format: VTT files; opens in any standard text editor
Transcript retention timing:
- Transcripts are stored for 2 years from the date of the appointment
- After 2 years, transcripts are automatically deleted
Delete transcripts: transcripts can be deleted through the Healthie API via the deleteTranscript mutation (learn more here)
Transcripts are not included in client EHI downloads
No Zoom cloud recording required for RTMS — cloud recording can be fully disabled
Scribe controls are appointment-type scoped: RTMS only runs when a provider with Scribe permissions books a Scribe-enabled appointment type. For appointment types without a Scribe template assigned, transcription does not run and clients see no notification

Staging note: AI Scribe and RTMS require Zoom to be enabled in Staging, which is not on by default. Testing in Production is recommended