🔊 Patter is the open-source way to connect any AI agent to real phone calls. Build voice-powered agents with a few lines of code — without managing telephony, speech pipelines, or audio infrastructure.
Visit the Patter docs to get started. SDK: Python / TypeScript
Patter is a dual SDK (Python + TypeScript) that gives developers a unified API for the entire voice AI pipeline — speech-to-text, AI reasoning, text-to-speech, and telephony — wired together with sub-second latency so your agent sounds natural on a real phone call.
import asyncio
from patter import Patter
async def main():
phone = Patter(
openai_key="sk-...",
phone_number="+15550001234",
webhook_url="your.server.com",
)
agent = phone.agent(
system_prompt="You are a friendly booking assistant for a dental clinic.",
voice="nova",
first_message="Hi! I'm calling from Smile Dental. How can I help?",
)
await phone.serve(agent, port=8000)
asyncio.run(main())🚀 Key Features
- 📞 Inbound & Outbound Calls — answer calls or dial out programmatically with Twilio or Telnyx
- ⚡ 3 Voice Modes — OpenAI Realtime (lowest latency), Pipeline (custom STT+TTS), ElevenLabs ConvAI
- 🔧 Tool Calling — let your agent check databases, book appointments, or transfer calls
- 📊 Built-in Dashboard — real-time call monitoring, cost tracking, and analytics
- 🧪 Test Mode — develop agents in your terminal, no phone needed
- 💰 Per-Call Metrics — cost breakdown by provider (STT, TTS, LLM, telephony) with latency percentiles
- 🛡️ Guardrails — content filtering and blocked terms before speech output
- 🏠 Local Mode — runs entirely on your machine, no cloud dependency
- 🤖 Any LLM — pluggable LLM loop, bring your own Anthropic, Gemini, or any provider
🏁 Getting Started
- 📥 Installation — install the SDK and set up your environment
- 📖 Documentation — full guides for Python and TypeScript
- 💬 Discussions — ask questions and share what you've built
- 🐛 Issues — report bugs or suggest features