Open source · Kubernetes · read-only

Know what happened
in your cluster.

OpsTide records what changed, what ran and what failed — and marks the stretches it wasn’t watching, so a quiet hour never reads as a calm one.

helm install opstide oci://ghcr.io/mihnk/charts/opstide -n opstide --create-namespace
Try the demo — no cluster needed
The OpsTide Explorer in the dark theme: a histogram of the last 24 hours, filters by record type, the list of records, and the panel of a Pod restarted after an OOMKill with its Before tab open, listing the Deployment spec write and the ConfigMap write recorded in the hour before it.

Evidence, not verdicts

It writes down what happened. Deciding why stays with you.

  1. OBSERVED

    Every record says how it is known

    Observed from the API, derived by a documented rule, or unknown. Nothing is dressed up as more certain than it is.

  2. Gaps are drawn, not hidden

    When it wasn’t watching — a restart, a missing permission — that time is hatched. A quiet stretch is never mistaken for a calm one.

  3. connection, not cause

    It links, it never blames

    It shows what changed before a failure — on the workload, its config, its node — and leaves the conclusion to you or your agent.

01 · Overview

Your day as a timetable.

Every CronJob run on one line, failures in red, what’s still due outlined. A five-day strip to scroll through time, the exceptions that need a look, and planned maintenance as a band — not as an alarm.

  • Missed runs and runs that ran long, grouped
  • Run time against its own baseline: “+34% longer”
  • Your time zone or UTC, in English or Turkish
The Overview in the light theme: counts for the last 24 hours, the watches that were not whole, a five-day strip of runs per hour with failures in red, and today’s timetable of CronJob runs.

02 · Explorer

Every record, one search away.

Filter like you would in your log tool — namespace:shop severity:warning — then open a failure and see what was recorded in the hour before it.

Recorded before this

  • 7 min · Deployment spec written by kubectl-edit
  • 12 min · ConfigMap it references changed, by argocd-controller
The Explorer in the light theme with two filters applied as tokens in the search bar, namespace shop and severity warning, a histogram of the last 24 hours and the matching records: a Pod restarted after an OOMKill, a Warning Event and a stretch that was not watched.

03 · Morning summary

The night, in your channel before the day starts.

At 08:30, or whenever you set, OpsTide posts what the night looked like to Slack, Microsoft Teams or a webhook of yours: failed runs, CronJobs overdue for a success, restarts with the change recorded just before them, and the stretches it was not watching. A quiet night says so.

  • Set up in two steps from the UI, which writes the Secret command and the chart values
  • The webhook URL stays in a Kubernetes Secret: never shown, never logged
  • Sent late if OpsTide was down at its time, never twice
The morning summary setup in the light theme: Slack, Microsoft Teams or a webhook to choose from, the time, how far back it looks and what it includes, beside a preview of the message built from the cluster’s record, with two failed runs, an overdue CronJob and a Pod restarted after an OOMKill.

What it keeps for you

  • Who changed it

    Every spec write with the client that made it — argocd, helm, kubectl — from Kubernetes’ own field managers.

  • Config changes, no values

    ConfigMap writes and the keys that changed, never their values. Secrets: metadata only, and only if you turn it on.

  • Releases and revisions

    Each rollout with its Pods, what changed in the template, and a fixed rule’s findings — not a health score.

  • Any moment, again

    Which image ran where last Tuesday at 10:00 — and whether OpsTide was watching then.

  • Scheduled jobs, kept honest

    Missed starts, run time against a 23-day baseline, and CronJobs overdue for a success — said only when it was watching.

  • Planned work, labelled

    Maintenance windows from your Helm values mark what happened inside them. Nothing is muted.

How it fits

Next to your stack, not in its way.

Kubernetes API

  • CronJobs · Jobs · Pods
  • Deployments · ReplicaSets
  • StatefulSets · DaemonSets
  • Events · Nodes · HPAs
  • ConfigMaps

OpsTide · one pod

Watches with list & watch, writes a timeline to SQLite on its own volume, serves the UI and a read API. Keeps 30 days by default.

token auth · /healthz · /readyz

  • Your channela morning summary on Slack, Teams or a webhook, if you want one
  • Prometheusoverdue CronJobs and its own health, for alerts
  • Grafanachanges & failures as annotations
  • Your agentversioned JSON evidence export

Security

Built to be allowed in production.

Read the full RBAC before you install. It is short.

Verbs
list · watch
Never asked for
exec, logs, Secret data, any write
Secrets
Off by default. When on: metadata-only requests, with the last-applied annotation and managedFields dropped on arrival. The grant itself is list & watch on secrets.
Scope
Whole cluster, or only the namespaces you list
Access
A token for the UI and API, or sign-in by name through oauth2-proxy
Outbound
Nothing, unless you set up the morning summary: then HTTPS to the destinations you list

Running in two minutes

Try it first · no cluster, no Docker

# sample data, opens on 127.0.0.1:8080
git clone https://github.com/mihnk/opstide
cd opstide && make demo-local

Needs Go 1.26, and Node 22 to build the UI.

Install in your cluster

helm install opstide oci://ghcr.io/mihnk/charts/opstide \
  -n opstide --create-namespace
kubectl -n opstide port-forward svc/opstide 8080:80
# then open http://127.0.0.1:8080

Sign in with the token from the opstide-auth Secret; every chart value is in the install guide. OpsTide is in early development (v0.1): the chart is published with the first release.