Project

2026Public

Persistent Open-World Simulation Engine

Project Brief

A state-management harness for long-running LLM simulation, maintaining player, NPC, quest, inventory, time, and world-state continuity across sessions.

Project Tech Stack

Core
TypeScript · React/Vite · Express
Interfaces
Gemini JSON generation · state delta updates · save/load flow
Artifacts
persistent world state · save snapshots · schema-driven outputs

Approach

The engine constrains model outputs into state deltas, tracks persistent entities across the world model, and records updates so each session can resume from an explicit game state rather than from conversational guesswork.

Memory scaffolding compresses prior events into recoverable world state, keeping player progress, NPC status, quests, inventory, and time evolution coherent over long interactions.

That makes the project a persistent-state prototype rather than a generic game demo: it shows how explicit state, schema constraints, and memory handling can stabilize long-running generative systems.

Outcome

It serves as a compact prototype for persistent state and memory scaffolding, demonstrating how recoverable world models can support long-horizon LLM interactions.

Media

Open-world simulation engine — gameplay demo
Open-world simulation engine — gameplay demo
Open-world simulation engine — LLM interaction
Open-world simulation engine — LLM interaction