Agent security path
Threat model → MCP checklist → audit trail → supply-chain kill-switch.
Guides
Secure AI infrastructure on Kubernetes.
Comparisons and checklists for GPU admission, MCP security, LLM observability, and serving stacks — written from production platform work.
01
Reading paths
Start with a route.
Ordered sequences across hubs — filter the library or open the next step.
GPU path
Pick a scheduler model, then put policy on the AdmissionCheck.
LLM serving path
Engine vs platform → scale on queue → budget before the GPU → observe.
Gateway planes
Split completion routing from tool admission; observe both.
GitOps path
Operating model first; promote only when policy can BLOCK.
02
Library
All guides.
Practical guides on GPU admission, BlastGuard kill-switches, MCP and agent security, LLM vs MCP gateways, inference budgets, Argo CD vs Flux, vLLM autoscaling with KEDA, and KServe versus vLLM.
AI infrastructure engineering
LLM Gateway vs MCP Gateway: boundaries and policies
LLM gateways govern model traffic; MCP gates control tool calls. Different boundaries and policies, with files from ai-runtime-platform and ASC.
OpenTelemetry LLM Observability: GenAI Conventions & vLLM
Practical OpenTelemetry for LLMs: GenAI semantic conventions, token usage metrics, vLLM Prometheus metrics and a token cost pipeline on Kubernetes.
Kubernetes GitOps architecture
Argo CD vs Flux in 2026: a hands-on comparison
Argo CD vs Flux on Kubernetes: reconciliation models, Application vs HelmRelease, self-heal and prune, plus kind lab timings from a public duel repo.
Agent and MCP security
AI Agent Audit Trail: What to Log per Tool Call
Record identity, policy, verdict, approvals and cost for every AI agent tool call. OpenTelemetry spans, digests and evidence chains on Kubernetes.
BlastGuard kill-switch: blast radius before promote
Blast-radius policy turns SBOM, VEX and exposure into ALLOW / WARN / BLOCK — fail closed when inventory is incomplete.
MCP Security Checklist for Kubernetes
A practical MCP security checklist for agents on Kubernetes: identity, token audience, tool pinning, egress, sandboxing, approvals and audit evidence.
OWASP Agentic Top 10 Mapped to Kubernetes Controls
Each risk in the OWASP Top 10 for Agentic Applications mapped to Kubernetes controls you can enforce: identity, admission, egress, sandboxing, audit.
LLM inference on Kubernetes
Inference Budget Guard: token budgets before the GPU
Per-team inference budgets on the gateway path: reserve estimated cost before chat completion, settle after, fail closed. Status from ai-runtime-platform.
KServe vs vLLM: Engine, Platform, or Both?
KServe vs vLLM on Kubernetes: what each layer does, how KServe runs vLLM, and when a plain vLLM Deployment is enough.
Autoscaling vLLM with KEDA: queue depth signals
Scale vLLM on Kubernetes with KEDA on Prometheus queue metrics: ScaledObject shape, vllm:num_requests_waiting, and how this relates to KServe autoscaling.
GPU scheduling on Kubernetes
GPU Admission on Kueue: fail-closed policy before the queue
How a Kueue AdmissionCheck enforces GPU flavors, FinOps labels, registry allowlists and soft quotas — with an evidence envelope per decision.
Kueue vs Volcano vs KAI Scheduler for GPU Workloads
Compare Kueue, Volcano and KAI Scheduler for GPU batch jobs: admission layer vs scheduler, gang scheduling, inference queues and when to use each.
Available for meaningful infrastructure conversations