Patter

Patter is an open source SDK that connects an AI agent to real phone calls. It handles the telephony work that is tedious to build by hand, including carrier signaling, two way audio streaming, turn detection, interruption handling, call recording, transfers and touch tone input, so that giving an agent a phone number takes a few lines of code. It is published as getpatter for Python on PyPI and for TypeScript on npm, under the MIT license. The SDK supports two architectures. In engine mode a single speech to speech model such as OpenAI Realtime or ElevenLabs ConvAI runs the whole conversation. In pipeline mode the developer composes speech to text, a language model and text to speech separately, choosing among providers such as Deepgram, Whisper, Cartesia, Soniox, Speechmatics and AssemblyAI for transcription, OpenAI, Anthropic, Groq, Cerebras and Google for reasoning, and ElevenLabs, Cartesia, Rime and LMNT for speech. Carriers include Twilio and Telnyx. The vendor cites response latency under 500 milliseconds, and a built in dashboard shows live calls, latency percentiles and spend. Patter is for developers building voice agents such as receptionists, appointment booking and outbound calling, who want to keep control of their own infrastructure. It is self hosted: it runs inside the customer's own environment, with no Patter account, no API key and no per minute fee from Patter on top of carrier and model bills. A hosted offering called Patter Cloud is listed as a waitlist, and the vendor offers demos. No prices are published for it.