KAPE — Kubernetes Agentic Platform Execution

Autonomous cluster monitoring, reasoning, and remediation — declared entirely in Kubernetes CRDs.

Why KAPE?

Kubernetes clusters generate constant signal — security violations, resource pressure, policy breaches, cost anomalies. Existing tools either act without reasoning (autoscalers, Kyverno) or reason without acting (K8sGPT). KAPE does both.

• Event-driven — ingests signals from Falco, Cilium, Kyverno, Karpenter, and K8s Audit via CloudEvents
• LLM-powered reasoning — LangGraph ReAct agents analyse context and produce structured decisions
• Declarative config — define agent behaviour entirely through KapeHandler, KapeTool, and KapeSchema CRDs — no code changes required
• Scales to zero — KEDA drives handler pods from NATS consumer lag; idle agents cost nothing

Quick Start

helm install kape ./charts/kape --namespace kape-system --create-namespace

Then declare your first agent:

apiVersion: kape.io/v1alpha1
kind: KapeHandler
metadata:
  name: falco-responder
spec:
  source: falco
  llm:
    model: gpt-4o
  tools:
    - name: kubectl-readonly
  schemaRef: incident-triage-v1

→ Full getting-started guide

Documentation

Architecture	How the platform fits together
CRD Reference	KapeHandler, KapeTool, KapeSchema fields
Design Specs	RFCs and detailed technical decisions
Contributing	Development setup and workflow

Status

Early development — core operator, runtime, and adapters are functional. Dashboard and task-service in progress.