Google Gemini Live

Google’s Gemini Live API runs a low-latency, native-audio dialog agent. The bridge proxies Teler call audio into Gemini and relays the model’s spoken response back.

Reference repos

Python teler-gemini-bridge

Node teler-gemini-node-bridge

Prerequisites

• A Google Cloud project with Gemini API access and a GOOGLE_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
GOOGLE_API_KEY	yes	must be provided	Your Google API key with Gemini access.
GEMINI_MODEL	no	gemini-2.5-flash-preview-native-audio-dialog (Python), gemini-3.1-flash-live-preview(Node)	Model identifier.
GEMINI_AUDIO_CHUNK_COUNT	no	20	Max chunks buffered before forwarding to Gemini — flushes immediately on generation-complete event
TELER_API_KEY	yes	must be provided	Your Teler API key.
NGROK_AUTHTOKEN	yes (local)	none	ngrok auth token.
SERVER_DOMAIN	auto	none	Public URL exposed to Teler. Detected from ngrok.

Tip

Gemini system message is the most direct lever to shape behavior: voice persona, tone, allowed topics. Iterate on it while developing.

5-minute setup

Python

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

Node

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

Initiate a call

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

Audio quirks

Gemini Live emits 24 kHz PCM. The bridge resamples to 8 kHz for Teler using scipy.signal.decimate in Python, and a chained biquad low-pass filter with decimation in Node.

Tuning latency

GEMINI_AUDIO_CHUNK_COUNT controls input buffering. Lower = lower latency but more network overhead and higher token consumption. Higher = smoother turns but laggier feel. The default of 20 is a reasonable starting point.

Troubleshooting

Symptom	Likely cause	Fix
Agent silent	Model name typo or unreleased	Check GEMINI_MODEL against current Google docs
API quota errors	Free-tier limits hit	Check usage in Google Cloud Console
Long latency to first token	Cold start	Send a warm-up message before connecting calls
Wrong language / persona	Empty system message	Set the system prompt explicitly

Next steps

Other recipes OpenAI Realtime, ElevenLabs, Vapi, and more.

Bridge architecture The conceptual model behind every recipe.