Orchestrating a Role-Playing Game Universe with Multi-Container AI Agents
summary
The talk describes building a containerized multi-agent RPG using small local LLMs (0.5B–15B) fine-tuned on ~6–12k QA pairs, combined with RAG and function-calling. The stack uses Docker Model Runner, a compose-aware model declaration, and MCP Gateway to expose tools, with Gen Kit (Go/Python) handling prompts and streaming; a terminal 4×4 dungeon crawler runs a master agent plus four NPC agents as MCP servers and is demonstrated live to completion.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 8:10Docker Model Runner for LLM
- 20:07Fine tuning small models
- 31:04LLM in Docker Compose
- 43:00Game AI with agents
- 1:07:35MCP protocol and servers
- 1:20:20MCP Gateway orchestration
- 1:36:00Summary
related talks
02h 48mCompose & Dragons: The Role-Playing Game of Agents Fed on Tiny Language Models
51mRolePl-AI
39mBuilding Agentic AI Apps with Docker MCP & Cagent
43mCoding a Multi-Agent Game Master with Spring AI
30mMind the Gap - A Developer's Roadmap To Building AI Agents
26mHow to Build and Govern Production-Grade Multi-Agent Systems
40mHow To Tame Your Agents?
18mBuilding AI Agents