Compose & Dragons: The Role-Playing Game of Agents Fed on Tiny Language Models
summary
Demonstrates building a text-based dungeon-crawler RPG driven by custom AI agents that use very small LLMs (<4B parameters). Uses Docker AI features and specific tiny models (examples: jan-nano-gguf 4B, qwen2.5 1.5B, granite-embedding 278M for RAG) to implement NPC personalities, a dungeon master for movement and combat, and dice rolls, and provides practical recipes and advice for reusing these techniques with tiny models.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:25Project origins and goals
- 17:06Docker Model Runner
- 29:14CLI demo
- 51:25NPC generation
- 1:08:05Context management
- 1:28:33Retrieval-Augmented Generation
- 1:44:53Room and dungeon generation
- 2:08:56Function calling and MCP
- 2:20:28Docker MCP Gateway
- 2:30:36Final RPG demo
related talks
01h 40mOrchestrating a Role-Playing Game Universe with Multi-Container AI Agents
01h 58mCompose & Dragons: the RPG of agents fed on Tiny LM
28mBuilding and Deploying A Multi-Agent AI D&D with Typescript
51mRolePl-AI
49mAI Apocalypse Survival Kit: coding agents & Tiny Language Models
44mGlimpses of Humanity: My Game-Building AI Pair
24mPlaying D&D With AI: Building an Agentic Game Engine
36mProduction-Ready AI Agents: From LLMs to Small Language Models