Adding auto healing for kubernetes

Right now, the kdm-cli is limited to monitoring Kubernetes and Docker environments, but it does not actively assist developers during pod failures or infrastructure crises. Instead of being just a monitoring tool, kdm-cli should evolve into a developer-support utility that can help diagnose issues, provide actionable insights, suggest fixes, automate recovery steps, and simplify troubleshooting during pod crashes, deployment failures, resource exhaustion, or cluster instability.

[Yuvraj-Sarathe]

@KDM-cli/core add your review on this

[Yuvraj-Sarathe]

We can make a tool that analyses error logs. It does not specifically need an LLM or API just some bunch of hardcoded suggestions can work like detecting file missing log error can just ask the user to handle the missing path file and etc.

Other commands like typo error or bad syntax in some file can also be assessed that way.

However, we cannot make an auto-correct engine or QnA type AI because that will require API keys and then keeping this free will not be possible.

Our options are limited and thus this might be the only way to upgrade

[utkarsh232005]

We can give options to them to add their api key for autohealing purpose which will store inside os keychain for security

[Yuvraj-Sarathe]

Well, still, I would prefer making a preset of data to feed into the AI about debugging because it will make it easier for them first, even without AI that can work for the most part.

Consider it like this: someone may not be comfortable enough to add an API call for an AI in his terminal during prod. But he still might need help; there we can make the pre-trained set of instructions to give and give him the answers to debugging.

Lastly, the AI we will be adding will not be as useful as one may think due to the restrictions of editing a file source. That AI also will just tell the debugging solution or point out that the syntax is not quite right; thus, the error occurred. This, to some extent, could be done without AI, but obviously AI would be better at that.

What it cannot do, though, is directly edit the file to solve the error because that will require giving it access to personal files that one may not want to give. Plus, we do not know which version they would be using, so a bad AI can break the code more than it can build.

The best we can do is make a preset of instructions that will be first fed into the AI before any prompt so it knows what the guidelines of production and deployment are. But another problem arises: CONTEXT LIMIT! Since this is such a big task, the pre-prompt might already fill it to the brim, let alone costing the API charges from the user!

My suggestion would be to develop such a function that figures out the problem via logs and tells the user what to do. Our breakthrough can be analyzing and catching errors before anything is broken as it runs simultaneously with the logs/workflow.

Hi @KDM-cli/core

We are officially kicking off the evolution of kdm-cli from a standard monitoring utility into an AI-assisted CLI tool for Docker and Kubernetes, heavily inspired by k8sgpt.

To ensure a smooth transition, this migration will be rolled out in phased add-ons, culminating in a fresh start for the v2 release.

I have just opened a PR for Phase 1, which introduces the foundational configurations required for this AI architecture. Please take a look and drop your feedback so we can align on the baseline setup!

[utkarsh232005]

all the discussion related to v2 kdm-cli will be discussed here

[Yuvraj-Sarathe]

I think you should assign each task to each member fir better results and also clearly mention targets because do we want a direct clone of k8s or a support tool for k8s (as the title of PR says for k8s) or a completely new program??

[utkarsh232005]

To keep things moving efficiently, I’ll be handling the core infrastructure setup solo for the first 8 phases. For now, please review the PR descriptions to understand the architecture. Once this baseline is solid, I’ll open phase-wise issues with clear targets and testing goals so everyone can claim tasks and contribute to the next rollouts.

[utkarsh232005]

🎉 Huge Milestone, @KDM-cli/core!

KDM now officially features integrated AI capabilities! The tool can now analyze Kubernetes environments and provide actionable resolutions for pod errors.

Step 1: AI Feature Testing

I will shortly share a set of testing .yaml files in a private repository.

Please clone the repository, run the configurations, and thoroughly explore the new AI features so we can gather feedback.

Step 2: The Next Roadmap (Interactive Dashboards via Phased Issues/PRs)

Once the AI baseline is solid, our next major goal is to convert our existing commands into interactive dashboards (similar to KDC) for a completely hassle-free user experience.

To keep this organized and manageable, every single command transformation will be handled via a separate tracking issue and a dedicated PR.

We will follow this exact workflow:

📋 Create a Dedicated Issue

Create a dedicated issue for a specific command.

Example:

Feature: Update kdm auth add command to be interactive

🛠️ Submit a Separate PR

Work on and submit a separate PR explicitly linked to that issue.

✅ Move to the Next Command

Once the PR is merged, move on to the next command.

---

Let's nail the AI testing first, and then I'll start opening the individual command issues so everyone can claim tasks! 🚀

[utkarsh232005]

@Yuvraj-Sarathe @Rishiraj-Pathak-27 just go through the current arch and test the commands after that share feedbacks in this discussion and after that we will start the step2