Hello. I have perhaps some naive questions but I n...
# ai
r
Hello. I have perhaps some naive questions but I need to start with something. Let's say I have "a system that can do things" and I have a Kotlin library (based on coroutines, so really usable only with Kotlin) with a number of functions to operate "the system". I would like to build an AI agent, which will be able to autonomously interact with "the system" based on commands given by the user, using different, configurable models as its "brain" (both paid or free, remote and local models should be supported). What Kotlin tools and libraries should I check? Is Koog the right tool for this? Should I write my own MCP server first?
f
I think Koog would be a great option to get started.
You can just define access to "your system" trough custom tools, switch the "brain" using Koog's providers. And weather or not you have an UI/API/CLI to allow the user to interact with the agent, you can just send prompts (independent of the transport layer) So yeah, you can "embed" Koog inside your existing system and expose only specific tools to it. And, even better, you can create custom strategies to be more deterministic. Hoe this helps 😄
thank you color 1