Deepgram

Deepgram is a streaming speech-to-text provider. This bridge wires Teler call audio to Deepgram’s real-time API; the transcript flows back into your application logic.

Note

Unlike the realtime-model bridges (OpenAI, ElevenLabs, Gemini, Vapi, Ultravox), Deepgram is STT only. You bring your own LLM and TTS. Use this recipe when you need fine-grained control over the conversation pipeline.

Reference repos

Python teler-deepgram-bridge

Node teler-deepgram-node-bridge

Prerequisites

• A Deepgram account with a real-time API key
• A Teler account with an active API key
• One Teler phone number on your account
• An ngrok auth token (for local development)
• Docker and Docker Compose

Configuration

Variable	Required	Default	Description
Deepgram_API_KEY	yes	none	Your Deepgram API key.
TELER_API_KEY	yes	none	Your Teler API key.
NGROK_AUTHTOKEN	yes (local)	none	ngrok auth token.

Caution

The variable name is Deepgram_API_KEY (mixed case). Match it exactly in your .env file.

5-minute setup

Python

git clone https://github.com/frejun-tech/teler-deepgram-bridge
cd teler-deepgram-bridge
cp .env.example .env
docker compose up -d --build

Node

git clone https://github.com/frejun-tech/teler-deepgram-node-bridge
cd teler-deepgram-node-bridge
cp .env.example .env
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"
  }'

How it works

The bridge opens a WebSocket to Deepgram’s real-time STT endpoint and pipes caller audio in. Transcripts come back as they’re produced, typically every few hundred milliseconds.

To build a full agent on top of Deepgram, fork this repo and add:

1. An LLM: call OpenAI, Anthropic, or your model with the transcript.
2. A TTS, such as Cartesia, ElevenLabs, or OpenAI, to produce audio.
3. Send the TTS audio back to Teler over the same WebSocket using the outbound audio message type.

Troubleshooting

Symptom	Likely cause	Fix
Empty transcripts	Wrong sample rate or codec	Check Deepgram model expects 8 kHz linear PCM
401 from Deepgram	Invalid Deepgram_API_KEY	Rotate; check casing in .env
Bridge doesn’t speak back	This is STT only	You must add LLM + TTS in your fork

Next steps

Pre-built voice agents If you don't want to build the full pipeline, use a realtime-model recipe.

WebSocket Messages The outbound audio messages you'll need to send TTS back.

Deepgram docs Real-time STT models and parameters.