Read-only
by default
An agent in the cluster and a read-only cloud role. Nothing is changed without an explicit, reviewable action — and platform vendors that provision for you need administrator rights to do their job.
Kubernetes · Cloud infrastructure · FinOps
Kubernetes, CI/CD and cloud infrastructure — operated, tuned and documented by engineers who measure the result. There is no platform to adopt, no rewrite, and we never ask for administrator credentials.
Services
Nine things, each one something we have done end to end. Engagements are scoped and priced before they start.
All servicesOperate
Build
Assess
Proof
Read-only
by default
An agent in the cluster and a read-only cloud role. Nothing is changed without an explicit, reviewable action — and platform vendors that provision for you need administrator rights to do their job.
−73%
on a claimed saving, after review
One recent audit: detectors summed to a headline figure, and reconciling overlaps cut it to a defensible range roughly a quarter of the size. We narrow the number before you see it, because a figure your engineer can break in one meeting is worth less than no figure.
How it runs
The same four steps whether it ends in a document or in an ongoing arrangement.
What you run, what hurts, and whether a diagnostic would tell you anything you do not already know. If it would not, we say so on the call and that is the end of it.
A read-only role and, where it helps, a metrics agent. Two weeks of measurement before any change, so the plan starts from what runs, not from what the diagram says.
You get: a read-only inventory and fourteen days of measured data
What we would change, in what order, what it costs and what we will not touch. Fixed scope, priced before it starts.
You get: a plan with a price and the order of work
A standing arrangement with a monthly written summary — or infrastructure as code, runbooks and decisions left in your repository so your team carries on without us.
You get: the monthly summary while we operate; at hand-over, infrastructure as code, runbooks and decisions in your repository
Fit
Three things make an engagement work. If one of them is missing we would rather say so on the first call.
Not for: application development and integrations, a round-the-clock operations desk, certification projects.
Not sure? Sixteen intake questions, most of them one clickEngagements
Four shapes, from a free fortnight of measurement to running the platform. Each one ends in something you keep.
Day-2 running of the clusters you already have: upgrades, capacity, autoscaling and the reliability work nobody schedules.
What you get
A platform that keeps working, and a record of why it looks the way it does.
Time and price: Month by month · a written summary each month · Either side can end it with 30 days' written notice.
The full review: the bill partitioned by owner, findings reconciled against each other, a plan in waves.
What you get
A written report you can forward, and a ranked plan with dependencies.
Time and price: Fixed price, quoted before start · fourteen days of measurement inside it
A specific question, answered. An architecture review, a second opinion, or an assessment before you commit a budget.
What you get
A decision you can defend internally, with the rejected options written down.
Time and price: By the day · a window agreed with the question
A written answer
A fortnight of measurement on your bill and your clusters, then a readout and a document you keep.
What you get
A measured baseline, a reconciled savings range, and a sequenced plan.
Time and price: No cost · about three weeks, of which fourteen days of measurement
Case studies
Anonymised under NDA, so the figures are generalised and the method is the part stated precisely. The method is also the part that transfers to your estate.
All case studiesWhy Kernforce
We are one of them, and not always the right one. Here is the honest shape of the alternatives.
Access required
What you keep afterwards
Pricing model
Who does the work
Lock-in
The number you are given
Our tooling
We develop MetrixForge, a platform for measuring Kubernetes cost, efficiency and risk: agents that collect metrics and scan images, an analytics layer over the result, and recommendations that can be applied as a reviewed pull request against your own repository.
By the numbers
Each of these restates something written in full elsewhere on the site. If one changes there, it changes here.
14days
of measurement before a number
0
administrator credentials requested
1
written summary of changes, every month
30days
notice to end ongoing operation, either side
Questions
A small team of senior engineers. The first call may be with a technical specialist, and the engineers on an engagement can change — someone is ill, someone leaves. That is why the work is written down as it happens, in your repository: the changes you reviewed, the runbooks and the decisions. Whoever picks it up starts from the record you already hold.
Upgrades planned against the provider's support window and rehearsed first, capacity and autoscaling reviews, an on-call path, and a monthly written summary of what changed and what it cost. Changes arrive as a pull request you review, not as a controller acting at 3am.
P1 — production down or degraded for users — is acknowledged within 30 minutes in working hours, 09:00–18:00 CET, Monday to Friday. P2 within 2 hours. Nights and weekends: P1 only, and only under a written on-call arrangement priced separately. Planned absences are announced in advance and written into the agreement.
No. There is nothing to migrate onto. We build MetrixForge and use it to measure, but the deliverable is a document and changes to your own repository — not a dependency on us. The agent is the one optional piece, and removing it removes nothing you keep.
Yes — that is the premise. Pipelines get measured before anything is rewritten, and GitOps delivery is set up so git keeps describing what actually runs. There is no platform of ours to migrate onto.
Writing
Written from engagements rather than from a content calendar.
All insightsA first call is a conversation, not a discovery questionnaire. If a short engagement is the right answer we will say so; if you do not need us yet, we will say that too.