Building AI Agents With Koog

Sebastian and Márton talk to Vadim Briliantov, the Technical Lead for Koog: an agentic framework built by JetBrains. We learn about what AI agents are, and why building them in Kotlin with Koog is a great choice. We also discuss all the different ways AI agents can connect to other systems and your existing code, and look at advanced features for agents like custom strategies, model switching, and history compression.

Resources:

• JetBrains/koog on GitHub
• Koog Documentation

Hosts:

• Sebastian Aigner – Website | Bluesky
• Márton Braun – Website | Bluesky

Guest:

• Vadim Briliantov – LinkedIn | Medium | GitHub | Bluesky | Twitter/X

Timeline:
(0:00) Intro
(1:11) Vadim’s history at JetBrains
(4:21) What’s an AI Agent?
(5:47) Koog!
(7:12) Applications for agents
(12:43) Koog’s building blocks
(15:05) Strategies, feedback loops
(23:55) The Kotlin DSL
(26:12) Persistent state
(29:48) Subgraphs
(32:33) Tools
(39:52) MCP support (and A2A)
(44:01) Entry point and type safety
(49:39) Spring and Ktor support
(51:27) LLM Providers
(53:30) Model switching
(56:02) History and memory
(59:22) Enterprise-ready
(1:02:12) History compression
(1:11:47) Markdown?!
(1:14:37) What’s next?
(1:18:22) Going open-source
(1:20:32) Conclusion