Inspiration

Twitch communities are built around shared moments, but those moments can lose their context when a streamer raids another channel. The receiving community suddenly has new viewers, new conversations, and new energy — but often no easy way to understand what brought those people together.

I wanted to build something that helps communities feel connected rather than making every raid feel like starting from zero. As a solo builder, I created THREADBOUND: Keep the community connected.

What it does

THREADBOUND is a raid-context bridge for Twitch streamers.

It captures a short, bounded window of chat before and after a raid, then uses the OpenAI API to create a concise context card highlighting conversation continuity, changes in discussion, and context that newcomers may have missed.

I designed the workflow so the streamer stays in control:

  • A pre-raid chat window is captured and frozen when a raid occurs.
  • A short post-raid window captures the receiving community's conversation.
  • With explicit AI consent, the bounded chat context is sent for analysis.
  • The resulting context card includes evidence references and supporting excerpts.
  • The streamer can edit, approve, or dismiss the draft.
  • Only an approved card is displayed in the OBS browser overlay.

When the available context is insufficient, THREADBOUND communicates that clearly rather than inventing a story.

How I built it

I built THREADBOUND independently, using OpenAI Codex as my coding partner. I used Codex to help plan the architecture, implement the Twitch integration and raid-event workflow, build the dashboard and OBS overlay, integrate AI-generated context, and develop automated tests.

The project uses Node.js, Twitch EventSub, SQLite, the OpenAI Responses API, and a browser-based overlay designed for OBS.

Privacy and safety were important design considerations. AI consent is off by default. Chat capture is bounded, viewer identity fields are discarded, and raw chat is kept in memory rather than stored as conversation history. AI requests use store: false and do not enable tools. Drafts remain hidden from the OBS overlay until I approve them.

Challenges and what I learned

One of my biggest engineering challenges was making AI-generated context useful and grounded in the conversation. I developed evidence checks that connect generated claims to source messages, alongside a streamer review workflow that keeps people in control of what gets shared.

Building THREADBOUND also taught me how to bring several systems together into one experience: Twitch authentication, real-time EventSub events, bounded chat capture, AI-assisted analysis, a dashboard, and an OBS browser overlay.

I created automated tests covering raid handling, privacy boundaries, evidence validation, the dashboard, and the approval workflow. The latest test run passed 61 tests, with zero failures and zero skipped tests. These automated tests use synthetic raid and chat data and mocked external API calls.

Building with Codex

As a solo builder, I relied on Codex to help me move from an idea to a structured application through iterative planning, implementation, testing, and debugging. I used it throughout the development process while making the product and design decisions myself.

THREADBOUND is my attempt to make raids feel less like a sudden arrival and more like a continuation of a community's story.

Keep the community connected.

Built With

Share this project:

Updates

Submission history