"Pretty dope!"
- Peter Steinberger (https://x.com/i/status/2079323356940403111)
Inspiration
Most AI experiences begin with the user prompting the model. The Guide reverses that relationship: you choose an experience, hand it to ChatGPT, and the story takes the lead.
I wanted to explore whether an AI conversation could feel less like operating software and more like stepping into an authored, interactive world — while still being transparent about what the experience will ask of the player.
What it does
The Guide is a cabinet of portable interactive-story cartridges called Elixirs.
A player browses the cabinet, previews an Elixir, downloads its self-contained Markdown file, copies the launch prompt, and opens both in ChatGPT. Before the experience begins, ChatGPT explains the game, its demands, capability boundaries, and data behaviour, then asks for affirmative consent and any required settings.
Once launched, the Elixir turns the conversation into an immersive choose-your-own-adventure experience. The model guides the player through the story, adapts to their responses, and stays within the cartridge's declared rules.
How it works
Each Elixir is a versioned Markdown cartridge containing:
- Structured metadata and a unique identifier
- The complete narrative and interaction logic
- Clear capability and data boundaries
- Consent and configuration steps
- Integrity markers so incomplete cartridges are not silently reconstructed
The Guide website provides the discovery and delivery layer: users can browse stories, preview what each experience involves, download a cartridge, and copy a safe launch prompt.
The cartridge format is deliberately portable. It does not require an account, database, API key, or custom runtime; the user's existing ChatGPT conversation becomes the runtime.
How Codex and GPT-5.6 were used
The Guide was built during OpenAI Build Week using Codex with GPT-5.6 across multiple development sessions.
Codex accelerated the project from an initial “reverse prompt” idea into a working product: shaping the cartridge architecture, building the responsive cabinet interface, creating validation and publishing tooling, refining the consent boundary, generating and editing story content, testing end-to-end flows, and preparing the GitHub Pages deployment.
GPT-5.6 was particularly valuable for maintaining coherence across product design, narrative design, safety boundaries, schema decisions, implementation, and testing. The core story-content session is supplied in the submission, while additional Codex threads were used for implementation and refinement.
Challenges
The central challenge was making plain Markdown behave like a trustworthy portable application. The format needed to be readable by people, reliably interpreted by a model, versionable in Git, and resistant to partial delivery.
Another challenge was balancing immersion with informed consent. The experience should feel magical after launch, but never conceal its demands or pretend that user-provided story instructions outrank the model's actual safety and capability limits.
Accomplishments
- A working mobile-friendly cabinet for discovering Elixirs
- A portable, versioned cartridge format
- Explicit pre-launch consent and capability boundaries
- Integrity checks for incomplete cartridge delivery
- A complete end-to-end experience requiring no custom backend or API credentials
- A reusable publishing approach for future stories
What I learned
Conversational interfaces can be authored as portable artefacts rather than hard-coded applications. The interesting design problem is not merely prompting a model; it is creating a dependable contract between author, player, and model.
I also learned that safety and immersion do not need to be opposites. A clear pre-flight explanation can make the fictional experience stronger because the player knows exactly when the role-play begins and where its boundaries sit.
What's next
Next I would expand the Elixir library, formalise the cartridge schema, add automated conformance testing, improve discovery and accessibility, and explore tools that let other writers create and validate their own Elixirs.
Log in or sign up for Devpost to join the conversation.