Kait (Kubernetes AI Tool)

kait is an AI assisted kubernetes debugging tool which automatically troubleshoots, diagnoses, and fixes issues related to kubernetes.

kait uses autonomous AI agents built using Microsoft's AutoGen

Installation and setup

You can install kait directly using pip:

pip install kait

kait requires an OpenAI API key which is read via the environment variable KAIT_OPENAI_KEY. You can provide a list of models to use and kait will use the available model. It is recommended to use models with the capabilities of gpt-4. Larger contexts work better too e.g. 'gpt-4-1106-preview'. Your environment variable needs to point to a list of models in the following format:

[
    {
        "model": "gpt-4",
        "api_key": "YOUR_OPENAI_API_KEY"
    }
]

Using a Local LLM

You can use OpenAI compatible local LLMs by including a base_url in your model spec:

[
    {
        "model": "chatglm2-6b",
        "base_url": "http://localhost:8000/v1",
        "api_key": "NULL", # Any string will do
    }
]

FastChat/ and llama-cpp-python both provide OpenAI compatible APIs which can be used with the above config. Which models provide adequate performance still needs validating.

Usage

kait requires kubectl to be installed and authenticated against the cluster you want to use.

To run kait simply run:

kait debug <DESCRIPTION OF THE ISSUE TO DEBUG>

For a full list of options, run:

kait debug --help

Examples

A number of examples/ are provided so you can see how kait performs.