Hi everyone,
long time forum user here. I am currently looking into various AI related projects and one of the things I am interested in are tool calling capabilities of some of the larger models.
The idea here is what would it look like if you gave an AI agent an interface for computer janitorial tasks. I am thinking about the possibility for the LLM to access certain files and the ability toi modify other, or to execute Ansible scripts.
I am also envisioning and event based massaging system. I am not sure yet what the best, Linux native way, would be to realize that, but for example that a cron job, or a program sends a message and then the AI agent becomes active.
My thought process here is that an AI agent with a rudimentary RAG system would have the ability to read logs instead of them needing to be parsed. At least for me a lot of automation that could be useful falls short because I would need to write logic that understands a certain programs file and log formatting structure.
The idea of this project is to build a PoC to see if this could work and to also build a test framework to benchmark how well modern LLMs would do in the year 2026 at these tasks to evaluate if they would be reliable enough to be of actual help for system administration or an unneeded overhead.