GhostCMD
AI DevOps that generates CI and deploys to Kubernetes
63 followers
AI DevOps that generates CI and deploys to Kubernetes
63 followers
GhostCMD is an AI DevOps engineer for indie hackers, founders and tiny teams. It connects to your GitHub repo, generates CI workflows, prepares Docker & Helm via an “AI Doctor”, and deploys your app to a managed Kubernetes cluster with HTTPS. Instead of wiring Actions, Dockerfiles, charts and TLS by hand, you plug in a repo and go from “GitHub repo” to “live HTTPS URL” without hiring a DevOps engineer.
@evgenyevgenyevgeny Congrats on the launch, Evgeny — “GitHub repo → live HTTPS URL on k8s” without reading Helm docs at 2am is a bold promise. I’d love to try this on a real side project. How do you handle secrets, rollbacks, and reviewing what the AI generated (PR/diff)? Any plan to deploy to a user’s own cluster too?
@hijacey Hey, thanks a lot for the questions.
Secrets
Right now secrets are managed per environment inside GhostCMD. They’re stored encrypted at rest and never logged. On deploy we sync them to:
GitHub Actions (as repo/environment secrets), and
Kubernetes Secrets in the managed cluster
so at runtime your app just sees normal env vars, and the raw values don’t leave infra outside of those two places.
Rollbacks
Deploys go through Docker + Helm, so every deploy is a new image tag + Helm release revision. Today rollbacks are pretty simple: you can redeploy a previous image/revision from the console if something breaks. A nicer “one-click rollback” from the UI is on my short-term roadmap.
Reviewing what the AI generates (PR/diff)
The AI Doctor doesn’t silently push to main. It works in a branch and opens a PR with the generated CI workflows / Dockerfile / Helm changes so you can review the diff, tweak anything you don’t like, and only then merge. If you don’t like it - just close the PR and nothing is changed in your repo.
User’s own cluster
At the moment GhostCMD deploys only to a GhostCMD-managed Kubernetes cluster (so I can control reliability and costs in the beta). “Bring your own cluster” is definitely planned - most likely via kubeconfig first, and then native EKS/GKE/etc integrations once the core flow is rock solid.
Tried GhostCMD this morning - really cool concept and the demo looks promising.
That said, a few rough edges I noticed:
• The onboarding still feels a bit “beta”: I had to jump between the app and the docs to fully understand the flow.
• Stack support seems limited for now (would love to see more presets beyond the current ones and clearer guidance for custom setups).
• The UI is functional but sometimes it’s not obvious what’s happening in the background during deploys / CI generation.
Overall though, I like the direction a lot. If you can smooth out the UX and expand stack/CI coverage, this could be super valuable. Congrats on the launch!