Summary

Patter is an open-source voice-AI SDK, offering a developer-centric alternative to Vapi and Retell. It enables AI agents to connect with phone numbers, handling telephony, STT, TTS, and real-time audio streaming. Available in Python and TypeScript with identical APIs, Patter allows developers to "own the stack," integrating with carriers like Twilio and Telnyx under an MIT license.

How It Works

The SDK bridges AI agents with the PSTN, abstracting voice infrastructure. Developers define agent logic; Patter manages STT, TTS, and audio streaming via WebSockets. It supports OpenAI Realtime for low-latency conversations and a Pipeline mode for granular STT/LLM/TTS control. Patter provides a unified API across carriers (Twilio, Telnyx) for features like call transfer and AMD, alongside vendor-neutral OpenTelemetry observability.

Quick Start & Requirements

• Installation: pip install getpatter (Python) or npm install getpatter (TypeScript).
• Prerequisites: Carrier (Twilio/Telnyx) and AI/voice provider (OpenAI, Deepgram, ElevenLabs) API keys are required, configured via environment variables.
• Setup: Minimal 4-line quickstarts exist for Python/TypeScript. Development uses an auto-provisioned Cloudflare tunnel (tunnel=True); production may need ngrok or static webhooks.
• Resources: Example templates and a PatterAI/skills repository are available.

Highlighted Details

• Carrier Parity: Unified API for Twilio/Telnyx, supporting DTMF, transfer, AMD, voicemail drop.
• Voice Modes: OpenAI Realtime (low-latency) or Pipeline (custom STT/LLM/TTS).
• Developer Productivity: Identical Python/TypeScript SDKs, concise quickstart, built-in tunneling.
• Extensibility: Supports custom LLMs, tool calling, and agent harness integration via skills.
• Observability: Integrated dashboard and OpenTelemetry tracing.

Maintenance & Community

Contributions are welcomed via pull requests and issues. Community channels and specific contributor details are not provided in the README snippet. The PatterAI/skills repo offers integrations for coding agents.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Permissive MIT license allows commercial use and integration into closed-source applications.

Limitations & Caveats

Production tunneling may require alternatives to the auto-tunneling feature. Specific provider configurations are necessary (e.g., Telnyx outbound profiles, ElevenLabs voice IDs). Whisper STT may require OpenAITranscribeSTT for production robustness due to potential minor hallucinations. Google Gemini free tier usage requires billing enablement for certain models.