Blog
From the blog
Practical Kubernetes guides, operations at scale, and notes from building KubeBolt.
Featured

Kobi and AWS DevOps Agent on the same incident
I removed a network rule from the database in my AWS lab and asked Kobi and AWS DevOps Agent the same question. Both found the cause and who caused it. What changes is where each one starts, how long it takes, what it costs and what it delivers.

The prompt carries the map, not the territory: 49,000 tokens of docs in 2,500
When you hand documentation to an assistant, the instinct is to paste all of it. The instinct is wrong, and the reason isn't context size — it's the cache. Kobi ships an index worth 5% of the corpus and fetches a whole page only when the question needs it. Here is what it cost, latency bill included.

Building an Autonomous Operations Engine: the 6-Layer Architecture
Most AI Ops tools are black boxes that return magic answers. Here's how we built the opposite: six layers with one job each, where the model proposes from a closed catalog and code decides what reaches your cluster.
More posts

What Autopilot reads in your cluster doesn't decide what it does
Autopilot reads annotations, logs and labels that anyone allowed to edit a workload can write. Since 2.6.6, that text counts as evidence, and only your KubeBolt settings decide whether it acts. We tested it with six attempts to steer it off course.

KubeBolt 2.6: requests from typical use, limits from the real peak
2.6 fixes the two numbers you size a container with: the spike a long chart never drew, and right-sizing that recommended with no data. How to read them now and how I use them.

Kobi has a face, and it doesn't smile
Kobi now has a face: a head that tells you what it's doing without opening the panel. Why it isn't a character, how to read its seven states, and what changed when we measured it at its real size.

KubeBolt 2.5: Kobi leaves the cluster, and you see what it reads as it reads it
KubeBolt 2.5 takes Kobi and Autopilot into your AWS account. Here's how an investigation reads now, how Kobi changes model when it leaves the cluster, and what each part costs.

Kobi on AWS: ten incidents and where each fix went
The AWS connector shipped in KubeBolt 2.5. Before that, we tested it with ten incidents in an EKS lab. This is how Kobi reads your account without writing a line of code, and why almost no fix was one more sentence in the prompt.

How I test Kobi in Azure: break something on purpose, then grade the answer
Before Kobi reads anything in your cloud, it has to find the cause when the cluster says everything is fine. We built a lab in Azure, break things the way they break in real life, ask Kobi the question an operator would ask, and grade every answer against criteria written before we start. This is the method, with what each test cost.

KubeBolt 2.4: every lit card says what's failing and since when
In 2.4, every lit card in KubeBolt names the resource that's failing, says since when, and takes you to it. Here's what each card says now, the color rules, what happens to a cluster when its agent stops connecting, and the rest of the release.

KubeBolt 2.3's new dashboards, card by card
In 2.3 we redesigned almost every KubeBolt dashboard. Each one now starts with a row of cards, each with its figure and its chart. Here's what every card on Overview, Fleet and Home tells you, when it changes color, and where to look next.

Connect Claude Code to your clusters, read-only
KubeBolt 2.2 introduces MCP-only API keys: they reach the MCP server and nothing else, with the role pinned to viewer and the cluster list you tick. Twenty-nine read tools for the agent you already have open, and thirteen of them are new.

A claim with no source isn't correct: it's unverified
Your code has tests. Your infrastructure has drift detection. The sentence on your site promising where your customer's data lives has nothing. We checked our own site against our own source and it didn't come through clean. Here is the method — and why the most dangerous finding is not the one that contradicts you.

Detection without memory is just noise: KubeBolt 2.1 gives insights a history
An insight that switches on and off can't tell you whether you have an incident or a pattern. In 2.1 every detection opens a durable episode with its own timeline, its recurrence counted and its policy per environment — and the product tells you when its own voice is off.

Three LLMs Are Better Than One: Cost Optimization in AI Ops
Using one big model for every incident burns money. Here's how KubeBolt splits the work across model-free rules, Haiku, Sonnet and Opus, and how we measure each stage before picking its model.

The Hidden Cost of Per-User Pricing in Observability
Most observability tools bill per active user and per host. The result: your monitoring bill scales with headcount, not with the system you actually run. Here's the math — and why per-resource pricing changes it.

Why We Built Another Kubernetes Tool in 2026
The observability market looks saturated — Datadog, Grafana, New Relic. So why build KubeBolt? Because there's a gap none of the incumbents can fill without rewriting their product.

Common Kubernetes Errors and How to Fix Them
A practical guide to the most common Kubernetes errors — CrashLoopBackOff, ImagePullBackOff, OOMKilled, Pending pods and probe failures — with the kubectl commands to diagnose and fix each.

We Built KubeBolt Open Source First
Open source isn't just a go-to-market for infrastructure tools — it's a covenant with your users. Why we shipped KubeBolt as Apache 2.0 and what we promise stays that way.

How to Automate Kubernetes Incident Resolution
Alerts tell you something broke — they don't fix it. A practical look at the levels of Kubernetes incident automation, from runbooks to autonomous remediation, and how to do it safely.
