Koog `0.5.0` is out with loooooooooooots of advanc...

# koog-agentic-framework

[Vadim Briliantov]

Koog 0.5.0 is out with loooooooooooots of advanced features 🔥
(should I put more “o” here?)

🤝 Make your agents connected with new A2A support. And by the way — Koog 0.5.0 features full-scale Kotlin Multiplatform A2A SDK available for all targets! You can use it with Koog agents or even without — it’s a library of its own (but of course, I hope you would use it with Koog 😉)

🗝️ Non-graph API for Agent Strategies is unlocked! Now you can experiment even faster and deliver custom strategies with plain Kotlin code while still having the access to most of Koog’s advanced features including state management and history compression.

🔙 Rolling back in time to a checkpoint using Koog’s Persistence now handles side-effects of all tool calls that already happened. Messed up your database state already? No problem! Koog allows you to register “tool rollbacks” for every tool, and it would use them to eliminate all side effects if your agent messed up — next step to enterprise readiness (and this feature is not present in any other AI framework among all platforms, including Python or TypeScript)

💿 Data types now fully define results of high-level components like subgraphWithTask or subgraphWithVerification — no need to implement finish tools anymore, Koog would generate everything automatically!

🛠️ Tools API gets a huge update: no need to construct ToolDescriptor manually even when using class-based tools on Multiplatform! Everything gets generated automatically, making it as easy as on the JVM (and my last article was already featuring this new API and usage examples)

👨🏻‍⚖️ LLM as a Judge component available out-of-the-box

🎏 Streaming now supports tool calls — highly requested feature…

And much much more!

Feel free to check out full release notes here: https://github.com/JetBrains/koog/releases/tag/0.5.0

[Michael Wills]

Definitely more "o"s needed 😅 🚀