prof-tournesol

Fix your Kubernetes errors with AI. 🌻

Goals

1. Connecting to a log system to detect Kubernetes errors
2. Attempting automatic error resolution via a model
3. If successful, automatically opening a pull request
4. If unsuccessful, making an AI-powered phone call to the on-call person
   • explaining the error verbally (the speech)
   • asking them to intervene
5. Logging the following information in a file:
   • time of the call
   • number called
   • speech used
   • response (boolean) to the question: can they intervene?

Architecture

How to use

Setup environment variables:

cp .env.example .env
vim .env

Setup your custom values with the credentials for your GitHub App + Discord bot + your Discord ID:

cp values.yaml.example values.yaml
vim values.yaml

Create a k8s cluster with our application + kubeai + k8sgpt + argocd:

just init

Add config files for a model (gemma3-1b) and k8sgpt to use it:

kubectl apply -f ./manifests/gemma3-1b-model.yaml
kubectl apply -f ./manifests/k8sgpt-config.yaml

You should wait for a while for the model to be ready, and everything to be up and running.

Once everything is ready, you can start the application that will fail:

kubectl apply -f ./manifests/argocd-hof-app.yml

If Prof. Tournesol (Bot) can solve the error, you will receive a pull request in your GitHub repository.

If Prof. Tournesol (Bot) can't solve the issue, you will receive an issue on your GitHub repository & an audio in your Discord DM where you can also respond with an audio.