Agent-Stream — Exotel Voice AI integrations

Sample and production-oriented bridges that connect Exotel AgentStream (bidirectional WebSocket audio) to Voice AI providers. Place outbound calls with the Connect Voice AI API.

Choose a provider

Provider	Path	Language	Best for
OpenAI Realtime	integrations/openai-realtime	Python	Speech-to-speech realtime
ElevenLabs	integrations/elevenlabs	Python	Conversational AI + ambience
Gemini Live	integrations/gemini-live	Node.js	Google Gemini Live
Sarvam	integrations/sarvam	Python	Indian languages (STT/TTS)
Pipecat	integrations/pipecat	Python	Pluggable STT → LLM → TTS (ExotelFrameSerializer)
Dograh	integrations/dograh	Python	Drop Exotel telephony into Dograh (platform wiring, not an AI vendor)
Agent recipes	integrations/agents	Python	Mix-and-match S2S + STT/LLM/TTS stacks on AgentStream

Agent recipe matrix (Exotel AgentStream)

Full picker: integrations/agents/README.md

Type	Examples
Speech-to-speech	GPT Realtime, Gemini Live, Gemini Live + Pipecat, Grok Voice
Pipelines	Deepgram / AssemblyAI / Sarvam / Speechmatics × GPT or Gemini × Cartesia / ElevenLabs / Grok TTS / OpenAI TTS

Quick start (OpenAI — still works from repo root)

python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt
cp env.example .env   # set OPENAI_API_KEY
python main.py        # shim → integrations/openai-realtime

Canonical OpenAI package: integrations/openai-realtime/.

Test with Connect Voice AI

1. Start a bridge and expose it (ngrok http <port> → wss://…).
2. Configure Exotel credentials (shared/env.exotel.example).
3. Place a call:

python shared/place_connect_call.py \
  --to +91XXXXXXXXXX \
  --stream-url "wss://YOUR_HOST/?sample-rate=8000"

Details: docs/CONNECT_VOICE_AI.md · WSS protocol: docs/AGENTSTREAM_WSS_PROTOCOL.md

Layout

shared/                 # place_connect_call.py + Exotel env example
docs/                   # Connect API + WSS protocol
integrations/<provider>/  # self-contained bridge + README + deps
integrations/agents/      # Voice AI stack recipes (S2S + STT/LLM/TTS)

Prerequisites

• Python 3.8+ (and Node 18+ for Gemini Live)
• Exotel account with Voicebot / Connect Voice AI enabled
• Provider API keys for the integration you run

License

MIT — see LICENSE.