Supercharging Spring AI: Scalable AI Agents with Koog + Spring AI + Spring Boot
summary
Demonstrates Koog integrated with Spring Boot and Spring AI to run durable, observable JVM agents: a banking example wires Spring controllers, Postgres chat memory and JDBC persistence, a multi-LLM prompt executor (OpenAI/Anthropic/Google), and type-safe subtask graphs that checkpoint execution with state-machine checkpointing so agents recover across crashes and machines. Uses history compression (50-message window and a fact retrieval compression strategy), benchmarks a 6–8% precision gain, and switches models on the fly (GPT-5, Sonnet) while recording telemetry.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 3:58AI agent system anatomy
- 7:11Spring backend setup
- 8:56Senses with Spring AI
- 10:01Agent brain with Cook
- 12:08Agent tools and workflows
- 14:52Chat memory for sessions
- 17:41Risks of LLM tool access
- 19:26Domain modeling for workflows
- 24:01Banking agent strategy
- 29:18Custom nodes and guardrails
- 32:31Fault tolerance with persistence
- 34:39Monitoring agents
- 36:02Cost optimization
- 40:31History compression strategies
- 43:43Cook integration summary
- 45:00Q&A
related talks
51mReliable AI Agents Using Domain Modeling with Koog in Java
01h 00mSpring AI
52mIntelligent Beans with Spring AI
55mSpring into the AI Era: A Practical Exploration for Developers
49mBuilding AI Agents with Spring & MCP
51mIntelligent Applications with Spring AI
27mBuilding AI Agents with Spring & MCP
54mBuilding AI Agents with Spring AI and Tanzu Platform