OpenAI Realtime

OpenAI’s Realtime API exposes a single WebSocket that handles speech-to-text, LLM reasoning, and text-to-speech in one pass. Pair it with Teler and you have a working AI voice agent in minutes.

Reference repos

Python teler-openai-bridge: FastAPI + websockets

Node teler-openai-node-bridge: Express + ws

Prerequisites

• An OpenAI account with Realtime API access
• A Teler account with an active API key
• One Teler phone number on your account (the from_number)
• An ngrok auth token (for local development)
• Docker and Docker Compose

Configuration

Variable	Required	Default	Description
OPENAI_API_KEY	yes	none	Your OpenAI API key.
TELER_API_KEY	yes	none	Your Teler API key from the dashboard.
NGROK_AUTHTOKEN	yes (local)	none	ngrok auth token for the public tunnel.
SERVER_DOMAIN	auto	none	Public URL exposed to Teler. Detected from ngrok automatically.

5-minute setup

Python

git clone https://github.com/frejun-tech/teler-openai-bridge
cd teler-openai-bridge
cp .env.example .env
# fill OPENAI_API_KEY, TELER_API_KEY, NGROK_AUTHTOKEN
docker compose up -d --build

Node

git clone https://github.com/frejun-tech/teler-openai-node-bridge
cd teler-openai-node-bridge
cp .env.example .env
# fill OPENAI_API_KEY, TELER_API_KEY, NGROK_AUTHTOKEN
docker compose up -d --build

Initiate a call

curl -X POST http://localhost:8000/api/v1/calls/initiate-call \
  -H "Content-Type: application/json" \
  -d '{
    "from_number": "+91XXXXXXXXXX",
    "to_number":   "+91XXXXXXXXXX"
  }'

The phone rings; the OpenAI agent answers. Talk to it.

Architecture

The bridge implements the standard 3-endpoint pattern documented in Media Streaming → Bridge:

Endpoint	Direction	Purpose
POST /api/v1/calls/initiate-call	client → bridge	Trigger an outbound call.
POST /api/v1/calls/flow	Teler → bridge	Returns a stream Call Flow.
WS /api/v1/calls/media-stream	Teler → bridge	Bidirectional audio.
POST /api/v1/webhooks/receiver	Teler → bridge	Lifecycle webhooks.

Inside the WebSocket handler, the bridge opens a parallel WebSocket to the OpenAI Realtime API and relays audio in both directions.

Audio quirks

OpenAI Realtime returns 24 kHz PCM. Teler operates at 8 kHz (or 16 kHz if you configure it). The bridge handles downsampling automatically; you don’t need to touch this unless you fork the repo.

Production deployment

• Replace ngrok with a real public host: Fly.io, Railway, ECS, or your own infra.
• Pin the Docker image with a tagged release rather than :latest.
• Set SERVER_DOMAIN explicitly in production; auto-detect is for local only.
• Forward Teler webhooks to your backend rather than the bridge for durability.
• Rotate OPENAI_API_KEY and TELER_API_KEY quarterly.

Troubleshooting

Symptom	Likely cause	Fix
Phone rings but silence	Missing OPENAI_API_KEY or invalid	Check container logs; recreate .env
Bridge can’t be reached	ngrok not running or SERVER_DOMAIN stale	docker compose restart; wait for ngrok
Garbled / chipmunk audio	Sample rate mismatch	Should not happen with this bridge; file an issue if it does
Long latency on first response	Cold start of Realtime session	Send a tiny warm-up message before connecting the call

Next steps

Other recipes ElevenLabs, Gemini, Vapi, and more.

Bridge architecture The conceptual model behind every recipe.

WebSocket protocol The exact messages flowing through the bridge.