RTC.ON NEWSLETTER – June 2026

We're 14 speakers deep and counting, and the program is already telling a story: Media over QUIC has gone from "interesting draft" to the thing half the room is shipping. Add a strong run of AI-in-live-production talks and some proper broadcast war stories, and the two days of talks in Kraków are filling up nicely. A few we're particularly looking forward to:

MoQ, straight from the source

Luke Curley, MoQ co-creator at moq.dev (ex-Discord, ex-Twitch)

Luke created Media over QUIC and wrote several of the core IETF specs, so this is as close to the source as it gets. His talk title is still under wraps, but he's also running a hands-on workshop on the 16th where you'll build a MoQ-powered audio/video call from scratch (and bolt on real-time speech-to-speech translation if you're quick). If MoQ is on your roadmap, start here.

Phone Cameras in Broadcast: From the Desert Stage to Millions of Viewers

Daniil Popov, Head of Technology at Cyanview

Cyanview's Elixir systems already run camera control at the Olympics and the Super Bowl. This time Daniil put a phone on the broadcast chain: a React Native Expo app wired into their RCP, plus a full 10-bit pipeline for iOS and Android. It ran at a major music festival this year, where a leading tech partner couldn't tell the phone footage from a broadcast camera.

Kyber: A Machine Control Stack Based on QUIC

Jean-Baptiste Kempf, CEO at Kyber

You know JB as the man behind VLC and VideoLAN. This time he's pointing QUIC at a different problem: machine control. Most of this year's QUIC talks are about moving media, and his is about moving machines, which makes it one of the more unexpected uses of the protocol on the program. The full abstract is still landing, but with that track record it's worth a slot on your schedule.

Computer Vision, Meet Sports

Piotr Skalski, Lead Open Source Engineer at Roboflow

Piotr walks through the full CV pipeline he built for basketball and football: detecting players, tracking them through occlusions, reading jersey numbers, mapping positions onto a 2D court, and computing live stats. Every model is open source. If you've ever googled your way out of a tracking bug, you've probably used something he made.

Also taking the stage:

More names are landing over the coming weeks. The full lineup, with talk abstracts and speaker bios, lives on the RTC.ON site.

Early Birds are now live, with limited amount available. Don't wait up, go get yours:

If half the lineup is talking about Media over QUIC, it's only fair we use it ourselves. As of Fishjam 0.28.0, MoQ streaming is live in the platform.

Quick refresher on why this matters: live video has always been a pick-two problem between scale, latency, and cost. HLS rides the CDN so it scales cheaply, but you're 10 to 30 seconds behind the camera (the reason your neighbour upstairs shouts "GOAL" before your screen catches up). WebRTC gets you under 200ms, but you pay for it with vendor-specific SFUs that don't scale like a CDN. MoQ uses the relay as a first-class part of the protocol, so one source fans out to many viewers like a CDN, but the relay pushes media the moment it arrives, keeping latency under a second. Sub-second latency at CDN scale, which is the thing neither HLS nor WebRTC can do alone. Karol's Livestreaming Trilemma post is the full story, and you can have a camera live on a Fishjam relay in about ten minutes.

We've also been pushing MoQ where most demos don't go: native mobile. MoQKit is our open-source native SDK with Swift APIs for iOS and Kotlin for Android. Not just playback, but publishing camera, mic, and screen capture, broadcast discovery, and raw data tracks for chat or telemetry, all over QUIC. React Native is next, built on top once the native foundation is solid. And to prove the point, we took MoQ Boy, the Game Boy-style demo that used to live in the browser, and ran it in a native iOS app: audio and video streaming out of an emulator, controls published back as a MoQ track, viewers fighting over the A button.

One nice bit of symmetry for the conference: Fishjam's hosted relay runs Luke Curley's open-source relay implementation, and Luke is the one giving the MoQ talk and running the workshop in Kraków. Same protocol, on stage and in the stack.

Want to try it? The MoQ tutorial is the fastest way in, and MoQKit lives at software-mansion-labs/moq-kit.

BOBDAHACKER | Link

I Could've Rickrolled the Entire FIFA World Cup. All I Needed Was My ID

Ever wondered how hard it is to take over a global broadcast, swapping the feed for your own right as Messi steps up for a penalty in front of millions viewers? Turns out, sometimes it's not that hard. The author of this article found his way into FIFA's internal systems and ended up with access to every single World Cup match stream, free to swap any live feed for whatever he wanted!

LIVEKIT | Link

Solving End of Turn Detection

Knowing when a user has actually finished speaking is one of the largest pain points in voice AI: cut in too early and you talk over people, wait too long and the conversation feels dead. At Software Mansion we know this one firsthand, our multimedia team wrestled with the same problem. LiveKit's new Turn Detector tackles it with some interesting ML classifiers that work straight off the audio, picking up the same timing and tone cues humans actually use instead of waiting on transcripts. The result is fewer false cutoffs across 14 languages, with an open-source mini variant that runs on CPU.

MEETECHO BLOG | Link

Can MoQ and WebRTC be friends?

Lorenzo Miniero explains MoQ to WebRTC nerds by mapping it onto familiar stuff: relays as SFUs, track aliases as mids, catalogs as a sort of SDP. The neat bit is the demo: a Janus plugin on his imquic library bridges both protocols, taking an OBS WHIP ingest and serving it to WebRTC endpoints and native MoQ subscribers at once. His take: the two can happily coexist, no killer narrative needed.

WEBRTCHACKS | Link

Vibe-coding WebRTC

Philipp Hancke describes how AI coding agents have become practical for work inside large, tightly-guarded codebases like libWebRTC and Chromium. He shares concrete examples from his own experience shipping libWebRTC improvements and features, and even implementing call-setup changes in Pion(WebRTC implementation) despite not knowing Go. His takeaway: AI doesn't replace the engineer, it lowers the cost of context-switching across the stack, deciding what to change still takes domain knowledge.