Features

One operator. Every cloud. Every stack you already run.

21 K8s probes (Sprint 2: NodePressure, DaemonSets, PendingPods, CrashLoopBackOff, ETCD, FailedMounts), 10 AWS + 10 GCP + 10 Azure cloud probes (live SDK + Cloud Monitoring / Azure Monitor wiring), four default fixers + one opt-in (all GitOps-aware), four AI tiers, and the safety envelope holding it all together.

K8s probes

shipped

21 native probes + 20 analyzers across 20 GVRs. Includes NodePressure, system DaemonSets, PendingPods, CrashLoopBackOff, ETCD, FailedMounts. Vault/ESO/CNPG/Ceph/cert-manager depth no SaaS tool matches.

AWS probes

shipped

10 probes covering RDS, EBS, EKS, IAM, ALB, ACM, KMS, S3, VPC. IRSA-first auth — no long-lived keys.

GCP probes

shipped

10 probes: Cloud SQL, Persistent Disk, GKE control-plane + node pools, IAM SA, Subnet, LB backends, Managed certs, GCS public-access, KMS. Live SDK wrapper + Cloud Monitoring API (disk-utilization on Cloud SQL). Workload-Identity auth.

Azure probes

shipped

10 probes: SQL DB (live storage_percent), Disk, AKS control-plane + node pools, Identity, Subnet (live IP-pool count), AppGW (live BackendHealth LRO), Certificates, Storage public-access, Key Vault. AAD Workload Identity native.

k3s

shipped

Srenix runs on any conformant K8s 1.27+ including k3s and RKE2. The same Helm chart, the same probes, the same fixers.

AI tiers (T0 → T3)

shipped (paid)

LLM-augmented enrichment, fix proposal, multi-step planning, and Vault runbook proposer — all approval-gated.

Ticketing integrations

partial

OpenProject in OSS. Jira and ServiceNow in the paid Enterprise tier. DriftReport-to-ticket idempotency built-in.

Action policy (the leash on the agent)

shipped

Operator-defined action_kinds, target-namespace allowlists, dual-approval thresholds, signed-JWT click-to-fix URLs, hash-chained audit. The brake is in code you can read before you install.

RAG learning loop + deep-RCA

shipped (paid)

RAG memory is live: Srenix reads prior resolutions before proposing and records every approve/deny/ticket-close outcome. Short-circuit default ON (skips the LLM when a high-confidence cleared prior exists). Paid opt-in: deep-RCA grounded in live web research via Firecrawl — a generic technical query is synthesized by the LLM (no namespace, hostname, or secret leaves the cluster) and the RCA is forwarded into every AI tier (T0 → T3). Pre-alpha v0.2.0-alpha.1.