Recipes

Recipes are reference integrations between Teler and popular AI voice providers. Each one is a runnable repo (clone, set environment variables, docker compose up) and ships with both a Python and a Node variant (except where noted).

Note

Most recipes follow the same architecture: a tiny Bridge service exposes /api/v1/calls/initiate-call, /api/v1/calls/flow, and /api/v1/calls/media-stream. The Bridge connects Teler’s WebSocket to the provider’s WebSocket and relays audio. See Media Streaming → Bridge for the full architecture.

Realtime model providers

A single WebSocket handles STT, LLM, and TTS. Lowest latency, fastest path to a working agent.

OpenAI Realtime GPT-4o realtime audio. Single WebSocket. 24 kHz output.

Google Gemini Live Native-audio dialog. Configurable model, prompt, and buffering.

ElevenLabs Conversational AI agents: voice, knowledge, and tools configured on ElevenLabs.

Ultravox Realtime voice model with low time-to-first-byte.

Voice agent platforms

End-to-end agent platforms that handle conversation logic on their side.

Vapi Vapi assistants: STT, LLM, TTS, tools, and routing.

LiveKit (outbound) Outbound calling into LiveKit rooms via Teler SIP trunk.

Speech-to-text providers

Bring your own LLM and TTS. These recipes wire Teler to a streaming STT provider only.

Deepgram Real-time STT. Best for English-first agents.

Devnagri Real-time STT optimized for Hindi and other Indic languages.

Common shape

Every WebSocket-based bridge follows the same pattern. Once you’ve run one, you’ve run them all.

git clone https://github.com/frejun-tech/teler-<provider>-bridge
cd teler-<provider>-bridge
cp .env.example .env
# fill TELER_API_KEY, NGROK_AUTHTOKEN, and the provider's API key
docker compose up --build

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

Standard environment variables across all bridges:

Variable	Required	Purpose
TELER_API_KEY	yes	Your Teler API key from the dashboard
NGROK_AUTHTOKEN	yes (local)	Tunnel for receiving Teler webhooks during dev
SERVER_DOMAIN	auto	Public URL exposed to Teler, set by ngrok
<PROVIDER>_API_KEY	yes	Provider-specific credential, see each recipe

Tip

In production, replace ngrok with a real public host (Fly.io, Railway, ECS, your own Kubernetes). Pin the bridge image, set health checks, and forward call.* and recording.* webhooks to your backend rather than the bridge for durability.

Build your own

The bridges are intentionally small. If your provider isn’t listed:

1. Fork the closest recipe (Deepgram for STT-only, OpenAI for realtime models, ElevenLabs for agent platforms).
2. Replace the provider WebSocket logic.
3. Adjust the sample-rate conversion if needed.

Or build from scratch using the SDKs: the StreamConnector primitive is exactly the abstraction the bridges use under the hood.