Inspiration
During a big Twitch collab, chat moves faster than anyone can read. Thousands of viewers across multiple channels are reacting, asking questions and starting inside jokes, and most of it scrolls away before a streamer ever sees it. The best moments for interaction get lost in the flood.
We built VibeGage to catch that energy and turn it into something streamers can act on and viewers can watch unfold live.
What it does
VibeGage gives collabs, co-op games, tournaments and community events a live read on the audience.
- Channel vibe check: summarizes each participating channel's mood, intensity and trending topics
- Question radar: groups repeat questions, shows how many people asked and identifies which streamer each question is meant for
- Chat vs. chat: surfaces where audiences agree and where they clash
A separate transparent OBS browser source turns it all into a playful on-stream scoreboard, with awards like Heaviest Motion (most chat messages), Heaviest Aura (most hyped) and Most Confused. Chat stops being background noise and becomes part of the show.
How we built it
VibeGage runs on Node.js, TypeScript and Fastify. Broadcasters authorize their own Twitch channels, and Twitch EventSub WebSockets deliver chat messages and Shared Chat participant updates. Every 15 seconds, VibeGage deduplicates messages, summarizes a rolling three minute window and pushes updates to the dashboard and overlay over WebSockets.
With an OpenAI API key configured, the app generates structured JSON summaries. Without one, a local heuristic summarizer and chat simulator keep the dashboard and overlay fully demoable, no Twitch credentials or API key required.
Challenges we ran into
Tone whiplash. Fast chats change mood in seconds, and reacting to every message made summaries jump around. A rolling window gave us the balance between staying current and keeping enough context to be useful.
Duplicates and crossed wires. Shared Chat produces duplicate message events, and viewers often ask a question in one channel that's meant for another streamer. We built deduplication and question routing into the pipeline so summaries stay clean and every question lands with the right person.
Accomplishments that we're proud of
We built a working end to end prototype that connects real Twitch chat to a streamer dashboard and an OBS overlay. The offline simulator lets anyone experience it with several realistic chats, even without Twitch authorization or a live channel.
Most of all, we're proud that VibeGage helps audiences feel seen. When you're one of thousands typing in chat, your voice rarely breaks through. VibeGage gives every viewer a way to shape the moment, while giving streamers real time cues and giving everyone a new visual moment to enjoy together.
What we learned
Context is everything in live chat. The channel a message came from, when it was sent and who it refers to can completely change its meaning. We also learned to design AI features with structured output and a local fallback, so the experience never breaks when an API key isn't available.
What makes VibeGage different
Tools like ChatCast already help a single streamer privately track the jokes, questions and topics in their own chat. VibeGage is built for something they aren't: collabs.
- Multi-channel by design: it reads every participating chat at once through Twitch Shared Chat, instead of one channel at a time
- Question routing: it identifies which streamer each question is meant for, even when it's asked in another streamer's chat
- Chat vs. chat: it compares audiences side by side, showing where they agree and where they clash
- Made for viewers too: the OBS scoreboard puts the audience's energy on screen, so chat becomes part of the show instead of a private dashboard
- Faster pulse: summaries refresh every 15 seconds over a rolling three minute window
What's next for VibeGage
- Testing VibeGage live during a high energy, multi-streamer collab. It has only been run during a stream where I was the only one in the chat and against simulated chats so far. Seeing it in the field would help evaluate how the summaries, question routing and scoreboard hold up under real chat volume
- Adding an on screen character who moderates the chat(s) and provides context updates on a rolling basis for viewers who entered mid chaos.
- Clip ready "moment" markers when chat energy spikes, so streamers can find highlights instantly
- Streamer controls to pin a trending question or award on screen
- Customizable overlay themes to match each creator's brand
- Post-stream recaps showing how each audience reacted across the event
Built With
- codex
- css
- dashboard-and-overlay-websockets
- fastify
- html
- javascript
- local-heuristic-summarizer
- node.js
- obs-browser-source
- offline-chat-simulator
- openai-api-with-structured-json-output
- rolling-three-minute-chat-window
- shared-chat-events
- twitch-eventsub-over-websockets
- twitch-oauth
- typescript