LiveKit (launched 2021) reimagined the open-source SFU around opinionated abstractions: a Room contains Participants, who publish and subscribe to Tracks. Signalling, server-side recording, simulcast, SVC, WHIP/WHEP ingress and egress, and SDK distribution for every major language are all built in. This is in contrast to mediasoup, which is more flexible but requires you to write more application code yourself.
LiveKit's break-out moment was OpenAI's adoption for ChatGPT Voice in 2024 — every real-time voice conversation through ChatGPT runs through LiveKit's audio infrastructure. Other large deployments include Snap, Reddit, Spotify and Asana. The managed cloud product (LiveKit Cloud) is a popular alternative to self-hosting for teams that don't want to run the SFU themselves.
Compared to mediasoup, LiveKit trades flexibility for time-to-market. Where mediasoup gives you an SFU and asks you to build the application around it, LiveKit gives you a fully-formed video API and asks you to use its abstractions. For most product teams the LiveKit abstractions are right; for teams with unusual requirements (custom routing, novel layered video) mediasoup is more flexible. The Go data plane is single-binary deployable, which makes Kubernetes-native deployments simpler than mediasoup's C++/Node.js mix.
