jev-got
Game of Thrones roleplay where an LLM writes each scene and Jev labels location, beat, mood, danger and whether prose stayed in fiction.
Overview
jev-got is a Game of Thrones roleplay in which you play Jon Snow. Each turn a narrator model (GPT-5.6 Luna via OpenRouter) writes the next scene, then Jev reads it and answers five questions in parallel: where Jon now stands (a location ID from a closed set), what kind of scene beat it was, its mood, how much danger he is in, and whether the prose stayed inside the fiction. Jev never writes anything.
The location header, soundtrack, background artwork and the next turn's prompt are all functions of those answers. If the in-fiction probability is below 0.5, the scene is regenerated once. Answers are stored so they can be re-resolved later without asking Jev again. Stories are kept in Redis, and it needs TypeSafe and OpenRouter keys.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Where does Jon now stand? | — |
| Choice | What kind of scene beat and mood was this? | — |
| Score | How much danger is Jon in? | — |
| Yes / no | Did the prose stay inside the fiction? | — |
How it works
- Player submits an action
- Narrator LLM writes the scene
- Jev answers five questions in parallel; regenerate once if out of fiction
- Update header, music, backdrop and next prompt from the answers
Strengths
- Game state derived from typed labels, not parsed prose
- Built-in check that the narrator stays in character
- Stored answers can be re-resolved without new calls
Caveats
- Needs Redis plus two API keys
- Fan demo using a commercial franchise setting
Getting started
Start Redis with Docker, `pnpm install`, copy .env.example to .env.local with TYPESAFE_API_KEY, OPENROUTER_API_KEY and REDIS_URL, then `pnpm dev`.At a glance
- Type
- App
- Status
- prototype
- Language
- TypeScript
- Runtime
- Next.js app with Redis
- Has code
- yes
Use cases and integrations
How curated lists describe it
Game of Thrones roleplay as Jon Snow. A story model writes the scene; Jev answers where he is, how much danger, and what should play under it.
AnotiaWang/awesome-decision-modelsA Game of Thrones text-adventure demo where a language model writes the story and Jev labels the scene.
logicrw/awesome-jev-projectsJev (TypeSafe AI) PoC through Game of Thrones.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
Location, scene beat, mood, danger level and whether the scene stayed in fiction.
Who writes the story?
A narrator LLM; Jev only labels scenes.
What happens if the narrator breaks character?
If the in-fiction probability is under 0.5 the scene is regenerated once.
Related projects
Video summarizer that ties every line to its second; Jev checks each summary sentence against what was said, with numbers checked in code.
ProjectbetaPythonClaude Code PreToolUse quality gate where Jev scores drafts (such as SVG illustrations) before they are written and blocks substandard work.
ProjectprototypeJavaScriptCompanion code comparing Google ADK and Microsoft Agent Framework for structured video timelines, with Jev as a shared per-segment QC gate.
LibraryprototypePythonEducational code where a local LLM voices a game NPC and Jev classifies player messages and checks every line for spoilers.
ProjectprototypePythonPlayful web app where twelve absurd characters vote on your idea via one Jev request with 36 independent questions.
DemoprototypeTypeScript3D Magic 8 Ball you shake and release, where Jev chooses among the 20 classic answers and shows a probability for each.
ServiceprototypeTypeScript