KAIF · Layer 1 · Executive Architecture

Kubernetes Agentic AI Fabric.

K · A · I · F — the blueprint for production agentic AI on Kubernetes

Three lifecycles — Development, Deployment, and Evaluation — unified by an AI Harness, running on a cloud-native Kubernetes substrate. Every capability maps to a CNCF or Kubernetes-native project — no parallel stack.

The 30-second view

One picture. Top to bottom.

Stakeholders consume agents. The Harness governs them. Three lifecycles produce them. Kubernetes runs everything.

Stakeholders
who consumes the agents
Product teamsEngineeringSRE / PlatformData & MLBusiness ownersEnd users
Agentic AI Harness
control plane · cross-cutting
Agent registryPrompt & tool catalogMemoryPolicies / guardrailsEval gatewayIdentity (SPIFFE)Observability fabric
Kubernetes
cloud-native substrate
Compute (CPU / GPU)Networking (Cilium · Gateway API)Storage (CSI · Object · Vector)Identity & RBACSecretsMulti-tenancy
Why this shape

What the picture says

Three lifecycles, one substrate

Dev, Deploy, and Eval are first-class citizens — each with its own CNCF tooling, all on Kubernetes.

The Harness is the seam

Prompts, tools, memory, policy and evals live in one control plane so agents stay consistent across teams.

Cloud-native by default

Every component has a Kubernetes-native equivalent — no proprietary lock-in, no parallel ops stack.

Where to go next

Progressively deeper layers