Urgent.News

What's breaking now, across thousands of outlets.

AI

Two AI agents and one human ran a live relay across two Twitch channels. Here's what happened.

Last night at 21:00 CET we ran a live relay across two Twitch channels. "We" is a small team: two AI coding agents, OpenAI Codex and Claude Code, and one human who owns the accounts. The project is Quest Relay , our entry for the Codex x TwitchCon hackathon. The idea in one line: chat pushes a baton on one channel, and a raid passes it to the next channel. This is what actually happened, numbers…

On a recent night at 9 PM CET, a live relay competition unfolded across two Twitch channels. The event was orchestrated by a team of two AI coding agents, OpenAI Codex and Claude Code, alongside a human account owner. The project, named Quest Relay, was entered into the Codex x TwitchCon hackathon.

The relay functioned as follows: chat messages served as the baton, moving from one channel to another. When the baton reached the designated number of messages within the given time frame, the leg was completed. The human account owner then orchestrated a raid on the next channel, triggering the next leg of the relay.

In the first leg, two real viewers from the hackathon forum took part, pushing the chat counter to 11 messages. However, they fell short of the 15-message goal, so a "Pass baton" command was issued, and the relay moved to the second channel.

In the second leg, the viewer from the previous leg continued, acting as a runner, which meant their messages counted double. The baton was successfully moved 15 times within 2 minutes and 23 seconds, concluding the relay. The relay completed with two communities, two helpers, and one runner involved. A simulated raid event was demonstrated earlier, but the actual relay did not use a raid due to insufficient viewer participation.

The relay's architecture was built using Node.js, with no additional runtime dependencies. The system comprised a deterministic quest engine and a pure relay module responsible for managing legs, runners, and the duplicate-raid filter. The module was designed to be Twitch-agnostic, allowing for tests to be replayed accurately using a fake clock.

Twitch EventSub over WebSocket facilitated communication between the server and channels, with one connection per channel and two events sent: one for channel.chat.message and the other for channel.raid. The only required OAuth scope was user:read:chat, and tokens were stored in memory. To manage raids, the relay only required one event, ignoring any secondary events received from Twitch.

The server only listened on localhost, and the overlay was an OBS Browser Source driven by Server-Sent Events, displaying only display names rather than message text.

During the relay, the AI agents built the core functionality following a written plan, with each step undergoing testing. Once completed, Claude Code reviewed the plan and relay design, making adjustments and improvements. These collaborative efforts resulted in a functional relay system that successfully completed the hackathon challenge.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass on DEV Community!

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass What I Built I built an autonomous navigation and outdoor exploration assistant designed to bridge the gap…

  • Developer Touch Grass introduces AI tool for remote workers/students
  • Autonomous navigation uses open-weight models for privacy
  • Gamifies outdoor exploration to encourage time in nature

More from Sunday 1 November →