It's powered by the whole ecosystem of KMP libraries which extend it with:
• automatic JSON schema generation for tool use / function calling, like Pydantic in Python and Zod in JavaScript, but much simpler to use and kotlinx.serialization native
• Automatic execution of tool use requests with error handling, so writing agents is as simple as putting messages += response.useTools() in a loop
• Media type detection based on file content simplifying attachment of non-textual content elements to messages
And many more. I use it everyday to power my coding and self-modifying agents, so I would say it's pretty stable, if it can analyse and modify its own code by adding tools. Kotlin is the best language to use with LLMs.
https://github.com/xemantic/xemantic-ai