Supercharging Spring AI: Scalable AI Agents with Koog + Spring AI + Spring Boot
Vadim Briliantov - JetBrains
AI agents are quickly becoming core components of modern applications, but running them in production introduces real challenges. Ensuring predictability, observability, fault recovery, and clean integration with Spring stack is essential for any serious enterprise usage on the JVM.
At JetBrains, we’ve faced these challenges head-on while shipping AI-driven features to millions of users. From day one we’ve been experimenting, failing, refining, and ultimately developing proven solutions for the real-world problems of AI agent scalability.
Here are just a few challenges we faced and solved:
- Operating AI agents at scale without sacrificing quality or blowing through budgets.
- Switching between available LLMs and tools efficiently on the fly without losing progress.
- Managing long-running conversations and avoiding token-limit failures.
- Building fault-tolerant agent workflows that recover gracefully inside Spring apps.
- Replacing brittle prompt chains with structured, maintainable workflows.
These lessons led us to Koog, an open-source JVM framework (for Java and Kotlin) designed for building reliable AI agents using a structured, testable, and highly observable workflow model. Koog integrates directly into Spring Boot and Spring AI, enabling developers to combine the best of all worlds:
- Spring Boot’s robust application lifecycle.
- Spring AI’s model flexibility and connectors and vector-store integrations.
- Koog’s orchestration, state management, resilience patterns and ready-to-use AI algorithms.
One of Koog’s defining capabilities is state-machine checkpointing: agents can persist and restore their full execution graph, not just a chat log. That means Spring applications can run long, multi-step processes that survive crashes, redeployments, or even move across machines. That is essential for real production systems.
In this talk you’ll learn:
- How Koog + Spring Boot + Spring AI combine to deliver observability and model flexibility.
- How to design AI agents as predictable, debuggable pipelines.
- How to build fault-tolerant, cost-aware, and testable agent logic.
- What we learned deploying AI-based features inside JetBrains products, and how to apply them directly in your own Spring systems.
Join us to see how AI agents on the JVM can become first-class components of modern Spring architecture.