Documentation

How Reflex works

Reflex is a Kubernetes controller that watches your Prometheus metrics, detects SLO breaches, and runs a chosen remediation. Three steps:

1. Detect. An SLO breach trips when a Prometheus query crosses your error-budget burn rate threshold over the configured window.
2. Analyze. The Recommender pipeline runs flap and blip filters, gathers Prometheus context (deploy age, traffic, memory, CPU), matches against curated patterns, and on Pro tier asks an AI reasoner for a root cause.
3. Act. Depending on the Reflex's automation level, the action either logs (observe), posts a Slack approval (dry-run), or executes immediately (auto). Every step is recorded to the local event log.

Trust ladder details are below; the live demo at reflexslo.io/demo shows the full loop in real time.

Requirements

• Kubernetes 1.24+ — runs on EKS, GKE, AKS, k3s, kind, and any conformant distribution.
• Helm 3.x — the official chart at charts.reflexslo.io is the supported install path.
• Prometheus 2.x — Reflex queries the Prometheus HTTP API. Bring your own, or bundle one with --set prometheus.install.enabled=true.
• Slack incoming webhook — optional, but recommended. Required for dry-run approval buttons on Pro.
• OpenAI API key — optional. Pro tier only. BYOK; the controller readsOPENAI_API_KEYat startup and never persists it.

Installation

Reflex runs in your Kubernetes cluster via Helm. Requirements: Kubernetes 1.24+, Helm 3.x, Prometheus.

Free Tier

# Add Reflex chart repository
helm repo add reflex https://charts.reflexslo.io
helm repo update

# Install (observe mode, 3 SLOs, 3 Reflexes)
helm install reflex reflex/reflex-runtime \
  --set prometheus.url="http://prometheus:9090"

Pro Tier

After purchasing at reflexslo.io/pricing, you'll receive a license key via email.

helm install reflex reflex/reflex-runtime \
  --set license.key="YOUR_LICENSE_KEY" \
  --set prometheus.url="http://prometheus:9090" \
  --set slack.webhookUrl="https://hooks.slack.com/services/..."  # optional

Quickstart: First SLO in 5 Minutes

# Port-forward Prometheus
kubectl port-forward -n monitoring svc/prometheus 9090:9090 &

# Preview what Reflex finds (dry-run)
reflex setup --dry-run --prometheus-url http://localhost:9090

# Apply SLOs when ready
reflex setup --prometheus-url http://localhost:9090

kubectl get slo -A
kubectl get reflex -A
kubectl get reflexrecommendation -A

kubectl logs -l app.kubernetes.io/name=reflex-runtime -f

Trust Ladder

Every reflex starts in observe mode. Promote when you trust it.

Promote a Reflex

# observe → dry-run
kubectl patch reflex <name> -n <namespace> \
  --type=merge -p '{"spec":{"automationLevel":"dry-run"}}'

# dry-run → auto
kubectl patch reflex <name> -n <namespace> \
  --type=merge -p '{"spec":{"automationLevel":"auto"}}'

Slack Integration

Reflex sends breach notifications and approval requests to Slack.

--set slack.webhookUrl="https://hooks.slack.com/services/YOUR/WEBHOOK/URL"

https://<your-reflex-controller>:8080/slack/callback

Configuration

Key Helm values for customizing Reflex:

Example: Availability SLO + Restart Reflex

# SLO: API must be 99.9% available
apiVersion: reflex.io/v1
kind: SLO
metadata:
  name: api-availability
spec:
  service: api-gateway
  indicator:
    type: availability
    metric: http_requests_total
    goodQuery: 'sum(rate(http_requests_total{service="api",status!~"5.."}[5m]))'
    totalQuery: 'sum(rate(http_requests_total{service="api"}[5m]))'
  objective:
    target: 99.9
    window: "30d"
  errorBudget:
    policy: high
    burnRate:
      critical: 14.4
      high: 6.0

# Reflex: restart when error rate spikes
apiVersion: reflex.io/v1
kind: Reflex
metadata:
  name: restart-api
spec:
  trigger:
    errorRate: "> 5%"
    window: "5m"
  action:
    type: restart
    target: deployment/api-gateway
  automationLevel: observe  # Start safe
  cooldown: "10m"

Free vs Pro

Troubleshooting

CRD Reference

After installing, all CRD fields are self-documenting:

kubectl explain slo.spec
kubectl explain slo.spec.indicator
kubectl explain reflex.spec
kubectl explain reflex.spec.action
kubectl explain reflexrecommendation.spec