Ralphthon

Track 1 — Codex Plugin

Build an extension people can plug into Codex and use immediately.

Create a Codex extension track submission: a Skill, MCP server, workflow, or tool package that other participants can install and try.

• • It reduces a real repeated workflow once connected to Codex.
• • Setup instructions, config, and README are clear enough for another builder to follow.
• • It is more than a prompt collection: it ships as a reusable tool, skill, or workflow.
• • The demo shows Codex calling the plugin and producing a useful result.

Examples: K-skill, an MCP for a SaaS tool, local file/browser/terminal tools, or a workflow package a team can reuse.

Think: any domain is fine. What matters is that it plugs into Codex and becomes usable immediately.

Track 2 — Image-native Service

Use GPT Image models as the core interface of a service.

Build a service where image generation or editing is not a side feature, but the main product experience.

• • Images are central to the input, output, editing loop, or app/game assets.
• • It goes beyond a one-shot image generator and creates a repeatable user flow.
• • Text, image, and UI connect into something people can save, share, edit, or play.
• • The use case avoids copyright, likeness, and prohibited-content traps.

Examples: generate game assets into a mini game, character/item cards, menu/poster/thumbnail tools, or image editing workflows.

Think: not a button that makes images, but an app that exists because images can be made and edited.

Track 3 — Busan Local Advantage

Use Busan as material, not homework.

Instead of forcing a local problem, use Busan/Gyeongnam assets such as ports, tourism, manufacturing, the sea, markets, and local culture.

• • The demo is more useful or more interesting because it is grounded in Busan.
• • It uses field-like inputs such as port/logistics documents, factory photos, market/tourism routes, or local language.
• • It picks a small situation a local user could try in a day.
• • Local context is connected to the product function, data, or UX, not just decoration.

Examples: multilingual port label/checklist tools, manufacturing SOP checkers, photo-based guides for Jagalchi/Nampo visitors, or games using Busan dialect and ocean imagery.

Think: not solving Busan as a problem, but using Busan as an unfair advantage.

Track 4 — AI Application

An open track for any application powered by AI.

If your build does not fit Codex Plugin, Image-native, or Busan Local Advantage, this is the open service track. Any app is welcome if AI is central to the product.

• • Text, voice, image, realtime, agents, search, automation, and multimodal AI are all allowed.
• • It needs a complete app flow where a user gives input and gets a useful result.
• • It should feel like a service with a problem, UX, and output, not just a raw model demo.
• • The strongest submissions create an experience that only became possible because of AI.

Examples: GPT Realtime voice input/output apps, meeting or call copilots, realtime tutors, multimodal search, agentic productivity tools, AI games, or social apps.

Think: if it does not fit the other tracks, that is fine. Build a useful or fun AI-powered app.

Track comparison

Declare your track at submission. Prizes are awarded per track.