Kubernetes incidents, resolved safely by AI.
Minh Phuoc Solution builds self-hosted AI Ops agents that detect, diagnose and remediate cluster incidents — with whitelisted actions, full audit trails and rollback built in.
From alert to fix, with a human-grade safety net
Every action passes through the same four stages, so automation never touches production without proof.
Observe
Informers watch Pods, Events and Nodes. A debounced pipeline filters noise.
Diagnose
A compact snapshot goes to a pluggable LLM that reasons about root cause.
Validate
Actions are checked against whitelist, owner-guard, bounds and cooldown.
Remediate
Dry-run first, then execute with an audit entry and one-step rollback.
Automation you can trust with the pager
AI agents are only useful in production if they are predictable. Guard rails are not an add-on — they are the architecture.
- Whitelisted actions only. Anything outside the allow-list is rejected.
- Bounded blast radius. Replica limits, owner checks, cooldowns and idempotency.
- Full audit trail. Every decision and change is recorded.
- Dry-run & rollback. Preview first, undo instantly.
Measured, not estimated — lower cost, faster responses and fewer hallucinations.
Tools built the way I'd want them if I owned the pager
Kite
Self-hosted AI Ops agent for Kubernetes with a custom CRD, a pluggable LLM layer and an auditable executor.
RunGuard
Runbook-driven incident remediation with dry-run mode, audit trails and rollback support.
Aura
Real-time async image moderation pipeline with a Next.js UI, Socket.io gateway and AI worker.
Infrastructure help for teams without a full SRE bench
Kubernetes & Cloud
Design, deployment and cost-tuning on AWS and DigitalOcean.
CI/CD & GitOps
GitHub Actions, ArgoCD and Helm pipelines with change-based deploys.
AI-native Ops
Controlled LLM agents integrated into your operations workflow.
Observability
Monitoring and alerting that cuts time to root cause.
Truong Minh Phuoc
DevOps engineer with a QA mindset: obsessed with failure modes, root causes and what “done” really means. Led DevOps and backend for a 5-service microservices platform running on Kubernetes at 99.9% uptime.
Let's make your on-call quieter
Tell me about your cluster and your worst incident. I'll reply personally.