Meet KITT

KITT (Kaleidoscopic, Interconnected Talking Transformer) is an example AI voice and video agent built on LiveKit, a real-time communications platform. It demonstrates how an AI assistant can join a live call as a regular participant — publishing its own audio and video tracks — and respond to spoken questions from the people in the room, including in group settings where users activate it with a phrase like "Hey KITT." Under the hood, KITT combines several services: Google Cloud Speech-to-Text turns spoken audio into text, an OpenAI GPT model generates a response, and Google Cloud Text-to-Speech converts that response back into audio, all connected over LiveKit's WebRTC infrastructure built on its Go SDK and the Pion library. The sample also shows live transcription on screen and shares transcription and AI-state data with everyone on the call, not just the person speaking. KITT runs through LiveKit Meet, a Zoom-style sample video application, and the underlying pattern can be adapted by developers building their own voice agents using LiveKit's SDKs. It is best understood as a reference implementation for LiveKit's real-time agent capabilities rather than a polished consumer product — a starting point for developers rather than something meant for non-technical end users. LiveKit itself is an open-source real-time video/audio infrastructure company with a commercial cloud offering.