kprompt is The AI Runtime for Kubernetes: it understands your cluster context and turns intent into operate / investigate / recommend — with a reviewable plan before anything mutates.
The stranger try path is not a chat REPL. It is a single broken rollout:
Observe → PlanResult-shaped proposal → approve → apply → verify → Learn Run it
# Install: https://kprompt.ai/docs/install kprompt demo git clone https://github.com/kprompt/kprompt-examples.git cd kprompt-examples && make magic-moment Needs Docker, kind, kubectl, make, git, and kprompt on PATH. Analysis is heuristic — $0, no provider key. The script prints the durable proposal before asking [y/N]. Non-interactive shells stop before mutation and print the exact apply line.
What you should see
| Beat | Honest signal |
|---|---|
| Break | checkout revision 2 stalls; revision 1 still serves |
| Propose | rollbackFailedRollout proposal, applied=false |
| Gate | Narrow policyAuto and your y / --approve |
| Learn | Patterns list after verify — outcome writeback |
Not auto-heal
| Claim | Reality |
|---|---|
| Silent prod fix? | No — propose → human approve → verify → Learn |
| Secret values? | Never |
| Learn = proof? | No — ranking / actionConfidence bias only (evidence-not-proof) |
Want the second hit (same fingerprint, visible Learn bias)? make learn-second-hit in the same repo.
Afterward
kprompt init --ollama kprompt "how's my cluster" Brownfield cluster? Prefer Adopt (bind-over-install). Advanced multi-break Observe fixture stays make walkthrough — not the first-run CTA.
Experimental — prefer kind / non-prod first.
Try: kprompt.ai · install · GitHub · examples
