Koog Comes to Java: The Enterprise AI Agent Framework From JetBrains | The JetBrains AI Blog

Koog for Java, an enterprise AI agent framework developed by JetBrains, integrates seamlessly with Java-based systems, allowing AI agents to be built directly within existing backends without the need for separate Python microservices. It offers a structured, fault-tolerant, and observable approach to orchestrating large language models (LLMs), making use of fluent builder-style APIs and native Java abstractions. The framework supports multiple workflow strategies, including functional, graph-based, and planning, to provide control over task execution order and tool availability, while ensuring fault tolerance through persistence features that save agent state to prevent progress loss. With integration capabilities for Spring Boot and support for major LLM providers like OpenAI and Google, Koog also enhances observability using OpenTelemetry, allowing for detailed monitoring of agent execution, token usage, and costs. Additionally, Koog offers history compression to optimize token usage and thread management for efficient resource allocation, making it a robust solution for incorporating AI agents in enterprise environments.