Open-source reasoning layer for Kubernetes: observe the cluster, plan safe actions, and execute only after approval. Laptop CLI + optional in-cluster Observe agent.
Start here → github.com/kprompt/kprompt · star · clone · try the demo
curl -fsSL https://kprompt.ai/install | bash
# or: brew install kprompt/tap/kpromptWebsite · Docs · Discussions · Examples
X · LinkedIn · Bluesky · hello@kprompt.ai
Kind cluster, seven broken workloads, heuristic Observe agent — zero LLM spend, Autopilot propose-only:
git clone https://github.com/kprompt/kprompt-examples.git
cd kprompt-examples && make walkthroughkprompt "scale api to 10" # plan first
kprompt "scale api to 10" --approve --wait # apply after review
kprompt "explain why payment-api is crashing" # debug
kprompt "delete everything in the cluster" # hard-denied
kprompt agent run -n payments --health --heuristicReads run immediately. Mutations show Intent / Plan / Risk / Actions / Blast radius, then ask y/N — or pass --approve.
Observe → Reason → Plan → Validate → Approve → Execute → Learn
| Step | What happens |
|---|---|
| Intent | Describe the change in plain English — or let Observe watch continuously |
| Plan | Typed actions, risk, blast radius |
| Safety | Wipe / delete-everything style prompts are hard-denied |
| Apply | Review, then execute against your local kubeconfig |
Same kubeconfig as kubectl. BYOK — your LLM keys stay local. Never silent auto-heal by default.
| Capability | |
|---|---|
| Plan → approve → apply CLI | Scale, deploy, rollback, Helm, explain, investigate |
| Observe agent | Namespace watch → Incident → gated Slack/webhook |
| Autopilot | Propose-only by default (never silent apply) |
| CI | --output json PlanResult for jq gates |
| Multi-context | Read fan-out + per-context mutate approval |
| Repo | Role |
|---|---|
| kprompt | CLI + Observe agent (start here) |
| kprompt-examples | Kind demos — make walkthrough |
| kprompt-website | Site + /install |
| homebrew-tap | brew install kprompt/tap/kprompt |
Website · CLI · Discussions · Good first issues · v0.7.0
X @kpromptai · LinkedIn · Bluesky · hello@kprompt.ai
Star the CLI · PRs welcome · The AI Runtime for Kubernetes.
