What is Patter Python SDK?

Patter is an open-source Python SDK for voice AI that connects any AI agent to real phone calls in 4 lines of code. It handles telephony, speech-to-text, text-to-speech, and audio streaming. Install with: pip install getpatter

Is Patter free to use?

Yes. Patter is MIT licensed and completely free. You only pay your own telephony provider (Twilio or Telnyx (Beta)) and AI providers (OpenAI, Deepgram, ElevenLabs) directly at their standard rates. No Patter account or API key needed.

How does Patter compare to Vapi, Bland AI, and Retell?

Patter is open-source and self-hosted, unlike Vapi, Bland AI, and Retell which are proprietary hosted services. Patter has no per-minute platform fees, no vendor lock-in, and supports any LLM. You own your data and infrastructure.

What voice modes does Patter support?

Patter supports three voice modes: OpenAI Realtime (lowest latency, all-in-one STT+LLM+TTS), Pipeline mode (mix any STT + any LLM + any TTS provider), and ElevenLabs ConvAI (ElevenLabs-managed conversational AI).

Open SourceMIT Licensed

Python SDK

Give your Python AI agent a phone number with 4 lines of code. Async-first, type-hinted, production-ready. No Patter backend needed.

pip install getpatter

main.py4 lines

1from patter import Patter
2
3# Connect to your telephony provider
4phone = Patter(mode="local", openai_key="...", twilio_sid="...", twilio_token="...")
5# Connect your agent to phone calls
6agent = phone.agent(system_prompt="Your prompt", voice="nova")
7# Make an outbound call
8phone.call(agent, to="+1...", from_="+1...")

Quick Start

4 lines to a phone call

Connect your telephony provider and start handling phone calls
Patter handles STT, TTS, telephony, and audio streaming
Embedded local server (FastAPI) — no Patter backend needed
Works with any LLM — OpenAI, Anthropic, or your own
Full type hints and async/await support

Getting started guide →

Features

Everything you need to connect your agent to phone calls

Three Voice Modes

OpenAI Realtime for ultra-low latency, ElevenLabs ConvAI for natural voices, or Pipeline mode with custom STT + TTS.

Telephony Providers

Twilio (mulaw 8kHz) and Telnyx (Beta, PCM 16kHz) out of the box. Bring your own credentials, no Patter account required.

Call Recording

Enable recording per call with recording=True. Recordings stored via your telephony provider.

Call Transfer

Built-in transfer_call tool. Your agent decides when to transfer — Patter handles the redirect.

Machine Detection

AMD on outbound calls with machine_detection=True. Auto-drop voicemail messages when an answering machine is detected.

Event Hooks

on_call_start, on_call_end, on_transcript — lifecycle callbacks for logging, analytics, and custom logic.

Embedded Dashboard

Built-in local dashboard with call metrics, history, and CSV/JSON export. No external tools needed.

DTMF Input

Keypad presses forwarded as [DTMF: N] in the transcript. Build IVR flows with your agent.

Dynamic Variables

{name} placeholders in system prompts replaced from a variables dict at call time.

Integrations

Works with your stack

Plug in any combination of telephony, speech-to-text, and text-to-speech providers.

Telephony

Twilio (mulaw 8kHz, auto-transcoding) and Telnyx (Beta, native 16kHz PCM). Inbound and outbound.

Speech-to-Text

Deepgram Nova (streaming) and OpenAI Whisper. Configurable language and model via phone.deepgram() / phone.whisper().

Text-to-Speech

ElevenLabs (streaming, multiple voices) and OpenAI TTS (24kHz, auto-resampled). Configured via phone.elevenlabs() / phone.openai_tts().

AI Models

OpenAI Realtime API, ElevenLabs Conversational AI, or bring any LLM via the Pipeline mode handler. Model-agnostic by design.

Start building in minutes

Install the SDK, write a handler, connect a phone number.

Read the docs →