Muhtalip Dede · Founder of kprompt2 min readBuilding AI SRE in Public
A long-form series on building an AI SRE under an approval boundary — intent compiler, PlanResult, safety, multi-context, investigation, and why we refuse unsupervised auto-remediation. Episode index and honesty rules.
kprompt started as an AI Kubernetes CLI: natural language becomes a reviewable plan, then you approve before apply. That wedge still ships today. The longer bet is AI SRE — a system that can investigate, explain why, show blast radius, and verify outcomes without silently mutating production. This series is how we build that bet in public.
We are not writing weekly changelog fluff. Each episode is a durable essay: a design claim, what already exists in the CLI, what is still building or exploring, and what we explicitly refuse. Share it on Hacker News or in a CNCF channel if the idea is useful even if you never install the binary.
Rules of the series
- Shipped / building / exploring — never imply a demo is a product
- Approval boundary stays load-bearing — no silent apply across contexts
- Typed outputs over chat vibes — PlanResult, risk, hard denies
- Prefer non-production while you learn; experimental software
- CLI stays Apache-2.0; no “buy Team to make the series real”
Episodes
| # | Topic | Status |
|---|---|---|
| 1 | Why AI SRE | Published |
| 2 | Intent Compiler | Published |
| 3 | PlanResult | Published |
| 4 | Safety Engine | Published |
| 5 | Multi-context | Published |
| 6 | Investigation Graph | Published |
| 7 | AI Timeline | Published |
| 8 | Cluster Memory | Published |
| 9 | Knowledge Graph | Published |
| 10 | Autonomous SRE — and why not yet | Published |
Read episodes in order: Why AI SRE → Intent Compiler → PlanResult → Safety Engine → Multi-context → Investigation Graph → AI Timeline → Cluster Memory → Knowledge Graph → Autonomous SRE (why not yet). Earlier positioning posts (PlanResult JSON field guide, vs kubectl-ai, plan-approve) remain companions; this series goes deeper and stays chronological.
Who this is for
- Platform / SRE engineers evaluating AI tools that touch clusters
- Builders designing agentic ops who need fail-closed patterns
- CNCF practitioners who care about GitOps, Prom, OTel — not only chat CLIs
Try the wedge while you read
Safe read on a non-prod context
brew install kprompt/tap/kprompt
# or: curl -fsSL https://kprompt.ai/install | bash
export KPROMPT_GEMINI_API_KEY="..."
kprompt "list deployments"
kprompt "optimize my cluster"Star the repo, open issues when an episode claims something the CLI cannot do yet, and follow the index as later episodes land.
Related posts
kprompt + kagent: PlanResult as an MCP tool under a CNCF agent platform
How to compose kprompt with kagent without collapsing the layers: kagent hosts Agents-as-CRDs via MCPServer / RemoteMCPServer; kprompt ships read/plan-only MCP tools that return a typed PlanResult and never auto-apply. Validated against kagent quickstart + first MCP tool docs.
Read articlekprompt on Google Cloud: GKE day-2 with Gemini, without a new control plane
Use kprompt against GKE the same way you use kubectl — get-credentials, aliases, plan-before-apply — plus Gemini BYOK. Optional Observe agent on the cluster; no Marketplace SaaS, no kubeconfig upload.
Read articlekprompt on AWS: EKS day-2 with BYOK, without a new control plane
Use kprompt against Amazon EKS the same way you use kubectl — update-kubeconfig, aliases, plan-before-apply — plus Ollama or cloud BYOK. Optional Observe agent on the cluster; no Marketplace SaaS, no kubeconfig upload. Native Bedrock preset still deferred.
Read article