This proposal introduces a framework for building AI-enhanced Decentraland scenes using a self-deployable AI orchestrator and supporting tools. Scene creators define capabilities that the AI can use, such as spawning objects, controlling NPCs, changing lighting, managing quests, or triggering scene events. The orchestrator connects player intent to these scene tools through an LLM.
The architecture follows an MCP-like pattern. The scene provides the tools, the orchestrator acts as the bridge, and the LLM serves as the reasoning engine. This allows creators to build dynamic AI-driven experiences while maintaining full control of their infrastructure.
Creators deploy their own orchestrator instance, manage their own API keys, and control operational costs. The orchestrator enables flexible AI behavior while remaining fully self-hosted and composable.
This grant delivers the orchestrator, a scene SDK, example AI-enhanced scenes, and developer documentation for $15,000 over 90 days.
Decentraland scenes today are largely static. NPCs follow scripted dialogue trees, events are pre-programmed, and player interactions follow fixed paths. This limits replayability and restricts what creators can build.
AI-driven environments change this. When an AI model can interpret player input and invoke scene capabilities in real time, creators can build experiences that were previously impossible:
The missing piece is infrastructure. There is no standardized way for AI models to interact with Decentraland scenes. This project provides that infrastructure.
A self-deployable AI orchestrator, a scene-side SDK, and a set of example scenes that demonstrate how to build AI-powered experiences in Decentraland.