The AI that fixes outages by itself — Patent pending
Talk to us ↗
Free now Early adopters: become a Design Partner — Limited places available 

Snapshot mode

Run an AI SRE against your cluster snapshot. No install. No RBAC. No changes.

You do not trust an unknown tool with your production cluster yet. Good — you should not. Point Srenix at a read-only snapshot — a directory of kubectl output you collected yourself — and read what it says. If the output is not useful, you close the terminal and nothing changed.

What it is

What snapshot mode is.

For platform, SRE and DevOps teams running Kubernetes in production. What you get is the same diagnosis engine that powers Srenix — 21 Kubernetes analyzers across drift, logs, workload and diagnostics — applied to a snapshot instead of a live cluster.

  • One command: srenix diagnose --snapshot <dir>
  • Input: a directory of plain, read-only Kubernetes output (what you export yourself).
  • Output: diagnoses — what looks wrong, why, with the evidence it used.
  • It cannot change anything. It was not given the ability. There is no install and no RBAC because there is nothing in your cluster doing anything.
  • Model: any OpenAI-compatible endpoint you choose, including in-cluster vLLM or Azure OpenAI; snapshot mode reads only the directory you pass.

Fifteen minutes

The fifteen-minute walkthrough.

Steps 1–2 are the commands you run yourself, outside Srenix, against a cluster you already have permission to read. Srenix only ever touches the local directory.

  1. 01

    Collect a snapshot (about 5 minutes)

    Export the read-only state you would look at in an incident anyway. A minimal shape:

    mkdir -p ~/sre-snapshot/$(date +%F) && cd ~/sre-snapshot/$(date +%F)
kubectl get deploy,sts,ds,job,cronjob,pod,svc,ing -A -o wide > workloads.txt
kubectl get events -A --sort-by=.lastTimestamp > events.txt
kubectl get pvc -A > pvcs.txt
kubectl get secret -A --no-headers | awk '{print $1,$2,$5}' > secrets-summary.txt
kubectl get nodes -o wide > nodes.txt

    Everything here is get — read-only. Add certificate-expiry and quota output if your cluster has them. The exact recommended file list ships with the tool; this page shows the shape.

  2. 02

    Anonymise if you want (optional)

    It is your machine and your directory: redact names or namespaces before you run anything. A directory of get output carries names, counts and shapes — not secrets or personal data — but the snapshot guide’s promise is that you control the file.

  3. 03

    Run the diagnosis

    srenix diagnose --snapshot ~/sre-snapshot/2026-11-02

    No install step in your cluster, because nothing is installed in your cluster.

  4. 04

    Read the diagnoses

    You get findings with causes and evidence — the kinds of incidents Srenix is built for: expired certificates, crash loops (CrashLoopBackOff), bad deploys, missing secrets, stuck jobs, full disks. Judge the reasoning on your own stack. That is the whole point.

    Terminal screenshots of a real run land here from the Content Marketer. None are shown until they do.

The honest list

What snapshot mode does not do.

  • It does not fix anything. Snapshot mode is diagnosis only. Fixes (policy-bounded, human-approved by default) need the installed product.
  • It does not watch anything. No triggers, no alerts — triggers (resource events, Alertmanager, HMAC webhooks) are part of the install.
  • It does not see live state between your exports. A snapshot is a photograph, not a camera.
  • Srenix is in beta, with pilots in progress; features, plans and prices may change before general availability. No certifications are claimed.

Where to go next

When you are ready for the next step.

Go deeper: how Srenix works · detect and verify · is Srenix legit · beta status · pricing.

FAQ

Snapshot questions.

Frequently asked

What teams ask before they point it at a cluster.

Something else? Ask us — a person replies.

Does snapshot mode need RBAC or a service account?

No. It reads a local directory you created. Nothing runs in your cluster.

Does my data leave my machine?

Srenix sends the snapshot content to the OpenAI-compatible model endpoint you configure — including a fully in-cluster or private endpoint. Nothing goes to a Reliora SaaS; there is no Reliora SaaS.

Which distributions work?

Anything you can run kubectl against; the installed product supports Kubernetes 1.27+ on EKS, GKE, AKS, k3s, RKE2 and OpenShift.

What can it actually find in a snapshot?

The analyzers cover drift, logs, workload and diagnostics across 21 Kubernetes analyzers; typical hits are certificates, crash loops, bad deploys, missing secrets, stuck jobs, full disks.

What if it is wrong?

Snapshot mode changes nothing, so a wrong diagnosis costs you only the read. In the installed product, every fix is bounded by policy, needs a human click by default, and is re-checked after.

Start with the snapshot. Install when you are ready.

This page is the snapshot step. When the output earns it, the install is one Helm chart — or we can walk a real incident on your stack with you.

Apply to be a Design Partner Book thirty minutes with an engineer