Declare intelligent agents as Kubernetes CRDs. They triage incidents, automate toil, and evolve from supervised to autonomous — so your SRE team ships reliability, not runbooks.
$ kubectl apply -f agent.yaml
The Problem
of pages are non-actionable. Every false positive interrupts deep work and degrades on-call quality of life. Existing tools generate alerts but can't determine which ones need human intervention.
senior engineers hold all the tribal knowledge. Runbooks go stale within weeks. When those engineers are unavailable, MTTR doubles and your team scrambles through outdated docs.
tool context switches per incident — Datadog, Splunk, PagerDuty, Jira, Slack. Every P1 means opening five tabs and mentally correlating data across systems. No single pane of glass exists.
Your team spends 40% of its time on toil instead of reliability work. Purko changes that.
The Solution
kubectl apply, manage with GitOps, govern with RBAC. Same workflow as everything else in your cluster.
apiVersion: agentic.io/v1alpha1 kind: Agent metadata: name: incident-triage-prod namespace: sre-agents spec: agentType: incident-triage autonomyLevel: "ha" model: provider: anthropic name: claude-sonnet-4-20250514 guardrails: maxActionsPerMinute: 20 requireApproval: false maxBlastRadius: "pod" rollbackOnFailure: true
Graduated Autonomy
Purko agents graduate from supervised to autonomous through the Shu-Ha-Ri model — inspired by martial arts mastery. Promotion is earned through measurable reliability.
Agent strictly follows predefined runbooks. All actions require explicit human approval before execution.
Agent adapts procedures based on context. Low-risk actions auto-approved; high-risk still requires human confirmation.
Agent operates autonomously within defined safety boundaries. Full self-healing with policy-gated execution.
Agent Catalog
Deploy production-ready agents in minutes. Each one is a CRD — version-controlled, testable, observable.
Correlates Kubernetes events, pod logs, and metrics to produce root-cause hypotheses within 90 seconds of alert.
haDetects anomalous log patterns using statistical analysis (z-score, frequency deviation) and flags emerging issues before they page.
haHandles CrashLoopBackOff, OOMKill, cert expiry, quota exhaustion — with policy-gated execution and automatic rollback.
shuUses historical metrics to predict resource exhaustion dates and recommends scaling actions with cost impact analysis.
shuAutomates cert rotation, node drains, config rollbacks, and scaling triggers with human-approval gates.
haAuto-generates incident postmortems from alert timelines, agent actions, and resolution steps. Exports to Confluence or Notion.
riBeyond SRE
Purko's agent archetypes — planner, executor, reviewer, router, monitor, and retriever — aren't limited to SRE. They map directly to every stage of the software development lifecycle.
The same CRDs, the same DAG workflows, the same Shu-Ha-Ri trust model. Define a code-development pipeline as an AgentWorkflow and Kubernetes orchestrates it — from requirements through production.
Extend in minutes, not months. Each agent is a CRD with a system prompt, model config, and tool bindings. Need a security scanner agent? A docs generator? A migration planner? Define the spec, apply it, done.
The framework is designed for CI/CD optimization, automated code review, dependency audits, release management, and compliance checks — all orchestrated alongside SRE agents in the same cluster.
apiVersion: agentic.io/v1alpha1 kind: AgentWorkflow metadata: name: code-development-pipeline spec: timeout: "45m" onFailure: stop steps: - name: plan agentRef: name: project-planner - name: implement agentRef: name: code-executor dependsOn: [plan] - name: review agentRef: name: code-reviewer dependsOn: [implement]
apiVersion: agentic.io/v1alpha1 kind: Agent metadata: name: code-reviewer spec: type: reviewer model: name: claude-sonnet-4-20250514 provider: anthropic temperature: 0.1 tools: - name: static-analysis type: builtin - name: vulnerability-scan type: api guardrails: maxIterations: 5 costLimitUSD: 3.0
Integrations
MCP-native connectors unify your tool stack into a single agent context. No custom integrations required.
Community-maintained connectors are open source. Enterprise connectors available with Pro and Enterprise tiers.
Purko is in early access. Join the waitlist to get notified when we launch.