Pipecat

Voice agents in Pipecat are built as composable pipelines: audio flows through a chain of frame processors covering speech-to-text, LLM inference, and text-to-speech, so each stage can be swapped independently. Maintained by Daily and licensed BSD-2-Clause, the framework integrates more than 20 STT providers (Deepgram, AssemblyAI, Whisper, Google), roughly 28 LLM services (OpenAI, Anthropic, Gemini, Groq, Mistral), and over 20 TTS engines (ElevenLabs, Azure, OpenAI), with transports spanning WebRTC via Daily or LiveKit, WebSockets, telephony, and WhatsApp. Client SDKs cover JavaScript, React, iOS, Android, and C++, and recent releases added multi-agent handoff, parallel fan-out, a debugging tool called Whisker, and a terminal dashboard called Tail. Installation is pip or uv (package name pipecat-ai), and a CLI scaffolds new projects. The framework is free and fully self-hostable, while Daily sells managed hosting as Pipecat Cloud; with about 14,000 GitHub stars it has become the de facto standard open source voice agent stack.