Exotel AgentStream: One repo for voice AI with Exotel

Exotel AgentStream: One repo for voice AI with Exotel

AgentStream is Exotel’s open-source repo for building voice bots that work with Exotel’s WebSocket streaming. You get both a Node.js multi-bot framework and a Python Gemini Live bridge in the same repo—pick the branch that fits you.


On main (Node.js)

One framework, 7 bots—run any of them with a single command:

Bot Command Latency
OpenAI Realtime npm run openai-realtime ~500ms
ElevenLabs Conversational AI npm run elevenlabs-bot ~750ms
Gemini S2S npm run gemini-bot ~2–4s
Gemini Live (beta) npm run gemini-live ~600ms
Pipeline / Simple Chat npm start, npm run s2s-bot ~4s

You get barge-in, 8k↔16k resampling, noise handling, Docker, and health endpoints.

Quick start: Clone → npm install → cp .env.example .env → add API keys → npm run .

Github: GitHub - exotel/Agent-Stream at nodejs-voice-bot-framework


On gemini-bridge (Python)

A small Exotel ↔ Google Gemini Live bridge: FastAPI + WSS, VAD/interrupt and Exotel clear handling, and a 10-minute setup guide with an architecture diagram.

(Use the branch dropdown on the repo and select gemini-bridge to see the Python code.)

  • Quick start: Checkout gemini-bridge → copy .env.example to .env → set your Gemini API key → run with uvicorn; follow the README and 10_MINUTE_GUIDE in that branch.

TL;DR

  • Same repo: * github.com/exotel/Agent-Stream .

  • main = Node.js framework (OpenAI, ElevenLabs, Gemini, pipelines).

  • gemini-bridge = Python-only Gemini Live bridge.

  • Switch the branch on GitHub to see the one you want; both are ready to run with Exotel’s Voicebot Applet.

1 Like