Skip to the content
dwim-ai
An AI watchman for Kubernetes

Your cluster breaks. We find out why, and propose the fix.

dwim-ai watches your Kubernetes clusters, investigates failures with AI, and shows you the exact change. Nothing is applied until a person approves it.

Sign inSee how it works
  • Read-only by default
  • Every change approved by a person
  • Full audit trail
Your clusterAgentDiagnosisCites evidence- memory: 128Mi+ memory: 256MiApproveApplied, checked and recorded
  1. 1A pod fails
  2. 2The agent notices
  3. 3AI explains the cause
  4. 4You approve the fix

How it works

From a failing pod to a verified fix

Four steps, and you are in the loop at the one that matters.

  1. 1
    Install one agent
    One Helm command. The agent only calls out to us, so you open no ports, and it can only read until you decide otherwise.
  2. 2
    Failures become incidents
    A restart storm, its crash loop and the rollout it blocks are grouped into one incident per workload, not a flood of alerts.
  3. 3
    AI finds the cause
    It reads status, events and logs through a short list of read-only tools, and every claim in its diagnosis cites the evidence behind it.
  4. 4
    You approve the fix
    You see the exact change as a diff. Approve it and the agent snapshots, applies it, checks it worked and records everything.

What you get

Built so you can trust what it does

Fewer alerts, an answer with its evidence, and a change you can read before it happens.

Catches what breaks
Crash loops, out-of-memory kills, bad images, stuck rollouts and pods that cannot be scheduled, noticed within moments and without noise.
One incident per workload
Related failures fold together. A workload that flaps stays a single incident with its full history, instead of opening a new one each time.
Diagnosis you can check
The AI cites the evidence it used, taken from your cluster. If the evidence is thin, it says so and lowers its confidence.
The exact change, as a diff
Before and after, side by side, checked against safety limits and the live state of your cluster just before you decide.
Approval that means something
An approval is tied to the exact parameters. Edit the proposal afterwards and the approval is cleared, so what you approved is what runs.
Safe by design
No shell and no kubectl. The agent runs a few typed, bounded actions, with limits, cooldowns and checks on its own side as well.
Roll back from a snapshot
The old spec is saved before anything changes, so putting it back is just another proposal, approved the same way.
Tell the right people
Alerts go to the contact points you choose, and roles keep it orderly: owners, admins who approve, and viewers who read.

Safety

You decide what it can touch

The agent starts read-only. Allowing changes is your choice, made in your own cluster settings, and it cannot be switched on from our side.

Read-onlyDefault
Watches, diagnoses and proposes. Has no permission to change anything.
Remediate
Runs only the actions you enable, only in the namespaces you allow, and only after approval.
  1. 1

    Kubernetes permissions. In read-only mode the agent has no permission to change anything.

  2. 2

    The agent's own gate. It refuses a change before it ever calls Kubernetes.

  3. 3

    Policy on live data. Limits are checked against your cluster as it is right now.

  4. 4

    Your approval. Nothing is applied until a person says yes.

AI on your terms

Spend AI where it helps

Each investigation uses credits from a monthly allowance. Run the AI automatically, only when someone asks, or not at all. If credits run out, only the AI pauses: detection, incidents, alerts and approvals carry on, and a diagnosis already under way always finishes.

When should the AI investigate?

  • Owners and admins can change it at any time
  • Usage is itemised: every run, its tokens and its credits
  • Credits are never taken for a run that failed before it began

Plans

Start small, grow when you need to

Online billing is on its way. Until then, plans are set up for you.

Free
Try it on a small setup.
  • 2 clusters
  • Incident detection and alerts
  • A monthly allowance of AI credits

By invitation for now

Pro
For teams running real workloads.
  • Up to 10 clusters
  • More AI credits every month
  • More AI runs at the same time

By invitation for now

Enterprise
Custom to how you work.
  • Custom cluster and member limits
  • Custom AI credits and terms
  • Your incidents go first in the AI queue

By invitation for now

Questions

Good things to ask

Stop chasing the same failures at 3 a.m.

Connect a cluster in a few minutes. It watches, explains and proposes. You stay in charge.

Sign in