// the take
A Next.js starter frontend for LiveKit's voice AI agents, wiring up LiveKit's JS SDK and 'Agents UI' component library into a working voice/video/screen-share session UI. It's for developers who already have (or are building) a LiveKit agent backend and want a frontend scaffold instead of starting from a blank WebRTC client.
The Agents UI components (control bar, session provider, chat transcript, five different audio visualizer styles) are genuinely functional, not just stubs - you get working audio-level visualization and session state handling on day one. Business logic is cleanly separated from UI primitives (components/app vs components/ui vs agents-ui), and because components are installed shadcn-style (copied into your repo, not hoisted into node_modules), you can actually edit the generated code instead of fighting an abstraction layer. Companion starters exist for Android, Flutter, Swift, and React Native, so a team can keep UI parity across platforms if they outgrow web-only.
Token generation ships dev-only — any client can mint a session token against the included route, and building the actual auth layer for production is left entirely to you with just a doc link, not code. It's a hard dependency on the LiveKit Cloud/Agents ecosystem specifically, so there's no path to reuse this if you ever move off LiveKit. Deep customization beyond Tailwind class overrides means editing or forking the vendored agents-ui source directly, since the components assume LiveKit's session object shape internally. There's no real test suite (CI just builds), and because it's designed to be cloned and modified rather than depended on, pulling upstream fixes into a customized fork will mean manual diffing, not a clean package upgrade.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →