send_transcript_to_openai_js

Send Transcript to OpenAI

This sample receives Zoom meeting transcript events through RTMS and sends each transcript segment to a configurable OpenAI model. It uses RTMSManager, the OpenAI Responses API, authenticated Zoom webhooks, and local usage controls.

Prerequisites

• Node.js 22 or newer
• A Zoom app with RTMS enabled and the meeting.rtms_started and meeting.rtms_stopped webhook events
• An OpenAI API key with access to the configured model
• A public HTTPS webhook URL for local development

Import manifest.json to create the user-managed Zoom General App. Replace the development and production domain placeholders first. The manifest requests only meeting:read:meeting_transcript and subscribes to the RTMS started and stopped events.

Setup

npm install
cp .env.example .env
node index.js

Set the Zoom webhook endpoint to your public URL plus WEBHOOK_PATH, for example https://example.ngrok.app/webhook.

Configuration

Variable	Default	Purpose
ZOOM_SECRET_TOKEN	required	Verifies URL validation and signed webhook deliveries
ZOOM_CLIENT_ID	required	Zoom app client ID used by RTMS
ZOOM_CLIENT_SECRET	required	Zoom app client secret used by RTMS
OPENAI_API_KEY	required	OpenAI API credential
OPENAI_MODEL	gpt-4.1-mini	Model passed to the Responses API
OPENAI_TIMEOUT_MS	20000	Timeout for each provider attempt
OPENAI_MAX_RETRIES	2	SDK retry limit for retryable failures
OPENAI_MAX_OUTPUT_TOKENS	512	Maximum generated tokens per response
OPENAI_MAX_INPUT_CHARACTERS	12000	Maximum request context size
OPENAI_MAX_REQUESTS_PER_MINUTE	30	Process-wide local request-rate limit; 0 disables it
OPENAI_MAX_REQUESTS_PER_STREAM	300	Request limit per RTMS stream; 0 disables it
OPENAI_MAX_SPEND_USD_PER_STREAM	1	Estimated spend limit per RTMS stream; 0 disables it
OPENAI_INPUT_COST_PER_MILLION_TOKENS	0.4	Input price used only for local spend estimation
OPENAI_OUTPUT_COST_PER_MILLION_TOKENS	1.6	Output price used only for local spend estimation
PORT	3000	HTTP server port
WEBHOOK_PATH	/webhook	Zoom webhook route
WEBHOOK_TIMESTAMP_TOLERANCE_SECONDS	300	Maximum signed-webhook timestamp age
MEDIA_SOCKET_CONNECTION_MODE	split	RTMS media socket mode

Provider pricing changes independently of this sample. If OPENAI_MODEL changes, update both cost variables using the current OpenAI API pricing so the local spend estimate remains meaningful. Provider-side quotas and billing limits remain authoritative.

Behavior

The sample subscribes only to RTMS transcript data. A verified normal webhook receives HTTP 200 before RTMS connection work starts. Each transcript event is sent as an independent OpenAI request, so conversation history is not retained.

Timeouts and retry limits are enforced by the OpenAI SDK. Local controls reject oversized input, excessive request rates, per-stream request counts, and projected per-stream spend before a request is sent. Actual token usage replaces the reservation after a successful response. Per-stream accounting is cleared when meeting.rtms_stopped arrives.

Provider errors are logged as sanitized fields with an error code, suggested action, retryability, status, and request ID when available. Response bodies and API credentials are not logged.

Webhook Security

Normal webhook deliveries are verified against the exact raw body using x-zm-signature, x-zm-request-timestamp, and ZOOM_SECRET_TOKEN. Missing, invalid, or stale signatures are rejected. URL-validation challenges are signed with the same secret token.

Docker

Build from the repository root because the multi-stage Dockerfile copies the shared JavaScript library:

docker build -f transcript/send_transcript_to_openai_js/Dockerfile -t rtms-openai-transcript .
docker run --rm --env-file transcript/send_transcript_to_openai_js/.env -p 3000:3000 rtms-openai-transcript

Runtime secrets are supplied through --env-file and are not copied into the image.

Files

• index.js: HTTP server, webhook verification, and RTMS event handling
• chatWithOpenAI.js: Responses API client and request controls
• .env.example: complete configuration template
• Dockerfile: platform-agnostic multi-stage Node.js image

See Zoom App Setup and RTMSManager documentation for shared setup.