Pipecat

Pipecat is an open-source (BSD-2) Python framework maintained by Daily.co for building real-time voice and multimodal AI agents. Its core abstraction is a Pipeline of FrameProcessors: audio frames stream in from a transport (Daily, LiveKit, Twilio SIP, WebSockets, or a phone bridge), pass through VAD and turn detection, optionally STT, then an LLM service (OpenAI Realtime, Gemini Live, Anthropic, Groq, Mistral, Together, and many more), then TTS (Cartesia, ElevenLabs, Deepgram, OpenAI), and back out the transport — with hooks at every step for tool/function calls, context management, and interruption handling. Pipecat ships dozens of service adapters for STT/LLM/TTS providers, an HTTP and WebSocket server harness, observability hooks (OpenTelemetry, Sentry), and a growing set of examples for telephony bots, browser voice agents, two-party translators, and avatar-driven assistants. Pipecat Cloud is the managed offering from Daily for deploying Pipecat pipelines on a serverless runtime without operating containers, with usage-based pricing on the underlying compute and any provider rates passed through. The combination of model-agnosticism, open source, and frame-level control has made Pipecat one of the most popular ways to build voice agents that don't lock you into a single vendor.