agents

Why we included this project

Most voice-agent demos stall at the notebook stage. LiveKit Agents is built for the next step: the agent runs as a realtime participant on your own server, composing speech-to-text, an LLM, and text-to-speech into one streaming pipeline. Semantic turn detection cuts interruptions, and built-in job dispatch routes end users to the right agent instance. The same agent logic reaches clients over WebRTC or places and receives phone calls through LiveKit's telephony stack. Agent code is written in Python (a JS/TS sibling library exists), and the framework works with most major AI providers, so you're not locked into a single vendor. Teams building call-center bots, telehealth assistants, or realtime translation tools where low latency matters more than a single API call will find most of the hard infrastructure already handled.