Detects workload problems, investigates them with your own LLM, and produces durable incident artifacts — all inside the cluster, with zero external dependencies.
Incident response is the bottleneck of platform engineering. SaaS AIOps tools solve it well — as long as your data can leave the cluster. For regulated teams, air-gapped clusters, and anyone with a BAA, HIPAA, PCI, or GDPR audit trail to worry about, those tools are off the table.
palena-sre is a Kubernetes operator that does workload-level incident detection and LLM-driven investigation entirely in-cluster. It doesn't host models, doesn't phone home, and doesn't force you to trust a vendor with production telemetry.
┌─────────────┐ ┌─────────────┐ ┌──────────────────┐
│ Detector │ ─────► │ Incident │ ─────► │ Investigator │
│ │ create │ CRD │ claim │ │
│ K8s events │ │ │ │ Evidence bundle │
│ PromQL │ │ Durable │ │ ReAct loop │
│ Metrics │ │ artifact │ │ Recommendations │
│ Logs │ │ │ ◄───── │ │
│ Alerts │ │ │ update └──────┬───────────┘
└─────────────┘ └─────┬───────┘ │
│ ▼
│ ┌──────────────────┐
│ │ Your LLM │
│ │ (OpenAI-compat) │
│ └──────────────────┘
▼
┌─────────────────┐
│ Notifications │
│ Slack / Teams │
│ PagerDuty │
│ Webhook / SMTP │
└─────────────────┘The architecture page walks through the full data flow.
kind cluster, point it at an Ollama or vLLM instance, and watch a crashlooping pod get investigated end to end. See the Quick Start.LLMEndpoint. The operator detects and notifies — LLM investigation is gated behind configuration you add later. See the installation guide.values-regulated.yaml preset, mirror images to your internal registry, and install into an air-gapped cluster. The installation guide covers it.Apache License 2.0. Copyright © bitkaio LLC.