Klastro is the Kubernetes lab platform we build at Edixos. Every lab is a real, isolated environment that opens in your browser, and the work you do in it is graded by reading the environment itself, not by checking what you typed. In November 2026 we took a fresh tour of the app: we ran a tutorial on a k3s node, repaired a broken two-node cluster in a graded challenge, opened VS Code and Argo CD inside a lab, and walked through the authoring, training and hiring screens. Every screenshot below comes from our own sessions.
Key takeaways
- Every Klastro lab is a real environment provisioned for you: a virtual cluster, k3s or kubeadm clusters with one or two nodes, or Linux machines. It is destroyed when the session ends.
- Grading reads the live state. Klastro checks whether the Deployment, the node or the file is actually right, so a correct answer reached your own way still passes.
- The same engine runs tutorials, graded challenges, open playgrounds, certification paths, instructor-led cohorts and proctored hiring assessments.
- Scenario Studio lets you author your own labs on a canvas and test them in a free preview, and an MCP server lets AI agents such as Claude Code drive the platform with your permissions.
- Klastro is in public preview, by invitation, with a free Community plan to start.
What is Klastro?
Klastro is a platform of hands-on cloud-native labs that run in the browser. Each lab gives you a dedicated environment, provisioned fresh and isolated from everyone else's, with nothing to install on your machine. When you finish a task, Klastro inspects that environment and tells you whether it is in the state the task describes.
That one idea sits between the two formats most engineers already know. A quiz checks that you remember the syntax of a manifest, never that you can bring a cluster back up. A throwaway demo terminal lets you type for twenty minutes and grades nothing. Klastro gives you the real infrastructure and a grade you can trust, and uses the same engine for learning, teaching and hiring.
Every scenario belongs to one of four types, which share the same environments but not the same rules:
| Type | What it is | Hints | Score |
|---|---|---|---|
| Tutorials | Guided, step by step, with a check on each step | Free | None |
| Challenges | A real task to solve on your own, graded when you submit | Cost points | Kept in your history |
| Playgrounds | A blank environment with no instructions and nothing to submit | None | None |
| Assessments | Exam material, run only inside a timed, proctored assessment | Cost points, cost hidden | Sent to the recruiter |
The scenario picks the machine it needs. Klastro offers a single-node virtual cluster, k3s and kubeadm clusters with one or two nodes, and Linux machines. On klastro.io, four of the five Kubernetes environments are described as giving each machine its own kernel behind hardware isolation, and every session gets its own network space, hosted in the European Union.
The dashboard adapts to your role
The dashboard is the page you land on after signing in, and it only shows what you can act on. A Needs you strip lists what is waiting, most urgent first: a candidate taking an assessment, results waiting for a verdict, a lab that failed. Next up puts your most recent lab one click away, with its progress, and a Practice record heatmap shows how regularly you practise.

The same page serves every role. A learner sees their labs and certification paths; an instructor sees live cohorts; a recruiter sees the latest results and an Invite candidate button. The shortcuts on the right follow the role and plan of whoever is signed in.
Find a lab in the catalog
The Learn menu opens the catalog: the official scenarios, grouped by topic and growing as the platform does, next to any your organization writes in Studio. You can switch between tutorials, challenges and playgrounds, search by name, filter by level and source, click tags, and sort by Fastest start.

That sort order matters. Some scenarios keep a pre-warmed environment ready, marked with a flame icon, and start much sooner. The documentation puts a pre-warmed Linux lab at 30 to 60 seconds instead of several minutes cold. In our session, a paused k3s lab resumed in well under a minute, and cold labs with setup steps took a few minutes.
Click Explore on a card and the scenario page tells you exactly what you are signing up for: level, estimated time, the environment, the list of steps or tasks, and the skills covered.

Inside a lab: instructions left, real machine right
Every lab opens in the same full-screen workspace. The left pane holds the instructions, a progress count and a Notes tab that saves as you type. The right pane holds terminal tabs: one per lab machine, plus a Shell tab with kubectl and helm already configured for the session. You can open up to five shells.
We started Under the Cluster: your first node, a tutorial that runs on a single-node k3s machine where you have root. The first step asks you to find the kubelet, which on k3s lives inside the k3s systemd service, and to write the unit's state to a file. We ran the commands in the terminal, clicked Check this step, and the step turned green.

A few details make the workspace pleasant to live in. Commands in the instructions are copy buttons. Hints, called guided steps in a tutorial, are free there. Each step allows a limited number of checks, three by default, which keeps you thinking rather than guessing. When you leave, Pause & resume later keeps the environment for another day, and End session destroys it for good.
Challenges: a broken cluster and a score to earn
A challenge drops you into a problem and grades the outcome. We picked The Agent That Stopped: a two-node k3s cluster comes up, then one of its nodes goes away, and the workload running on it is stuck. Four tasks, 55 points, and hints that cost points: the first one is marked −2.
The scenario reads like an on-call shift. kubectl get nodes showed node-2 going NotReady. The replacement Pod stayed Pending, and its events gave the reason: the Deployment's topology spread constraint would not put two replicas on one node.

On node-2, a systemd drop-in was pointing the k3s agent at a server address that never answered. We removed it, restarted the agent, the node came back Ready, and the second replica landed on it.
Then Finish & submit. Klastro checked every task against the live cluster and opened the results page: 4 of 4 tasks passed, 50 of 55 points. Points are earned check by check, so a task can pass without its full score (our second task earned 10 of its 15 points), and every graded run stays in My results after the environment is gone.

More than a terminal: editor and app views
Some scenarios add a Code tab, a full VS Code editor in the browser that works on the same files as your shell, and app views: web tools running inside your lab, shown as tabs next to the terminals. The catalog includes playgrounds built around tools such as Argo CD and Grafana. When a tool needs a login, the lab hands you the credentials in a strip above it.
In the Argo CD playground, we wrote an Application manifest for the classic guestbook app, applied it from the shell, and about a minute later it reported Synced and Healthy. The manifest sat in the editor the whole time.

Open in new tab gives the editor or the tool a full browser tab, still wired to the same lab.
Certification paths: CKA, CKAD and CKS practice on real clusters
Certification paths turn the catalog into a plan. Each path is a sequence of steps, from the first concept to timed mock exams, for Kubernetes and Linux certifications such as the CKA, and the list grows with the catalog.
Open a path and it lays out the route. The CKA path groups its steps into chapters, from architecture and workloads to networking, storage and troubleshooting on real kubeadm clusters, with milestones that track your progress along the way.

Steps unlock in order, a launch button opens the matching lab, and finishing a tutorial from a path ticks the step for you. Paths are independent practice: they prepare you for practical exams by grading real cluster state, and are not affiliated with or endorsed by the certification bodies.
Scenario Studio: build your own labs on a canvas
Scenario Studio is where instructors and teams write their own scenarios, without writing a test harness. A scenario is a graph on a canvas. You place an Environment, chain setup steps (a Helm release, manifests, a shell script, OS packages, files, a wait, an HTTP call), then add Tasks with their checks.

A check can read one field of one Kubernetes resource with JSONPath and judge it with a CEL expression, for example "three replicas are ready". Check lints the canvas, Preview launches the scenario on a real environment, and Test validation runs a task's checks against that preview so you can see them fail on the untouched cluster, then pass once you do the work. Previews cost no credits. Publish puts the scenario in your organization's catalog with a Studio badge.
Tracks and cohorts: deliver a class
For instructor-led training, a track is a reusable curriculum: an ordered list of scenarios, from the official catalog or your own Studio, plus materials such as slides, videos or links. You build it once, then deliver it to a group as a cohort.

Composing a cohort copies the track, so improving the track later never rewrites a class that is already running. During the session the instructor follows every trainee's progress, moves the class from one scenario to the next, and trainees can raise a hand without leaving their lab. This is the format we use for our own Kubernetes and platform engineering training.
Hiring assessments: a practical test with a report you can defend
The same engine turns into a technical hiring tool under an Enterprise contract. A recruiter composes an assessment template: a total duration, a passing score, and an ordered set of graded scenarios picked from the catalog or written in Studio. The time allocation bar shows how the scenarios fill the slot.

Candidates receive an invitation and work under a global timer, with live supervision available to the recruiter. Hints still cost points, but the candidate does not see the cost or the score. When the assessment ends, the recruiter gets a report: score against the passing mark, hands-on time, tasks passed, and a breakdown per scenario. The report below comes from a separate demo assessment with a 60% pass mark, and we blurred the candidate's identity.

Further down, a skill radar and a list of strengths summarise performance by area, and each task expands into its individual checks. The report can be exported as a PDF, and the recruiter records a verdict that shows on the dashboard.

Because every score comes from validation rules run against the real cluster, two candidates who sat the same scenario are compared on the same checks, task by task.
Drive Klastro from your AI agent
Klastro ships a Model Context Protocol (MCP) server, so an AI assistant can do in plain language what you would otherwise do in the app. The dashboard card gives you the command to copy. For Claude Code it is one line:
claude mcp add --transport http klastro https://mcp.klastro.io/mcp 
There is no API key to paste. The first call opens a Klastro sign-in and a consent screen, where you choose the organization and the permissions to grant: Read to look, Write to act. The agent then acts with your rights and nothing more: it can browse the catalog, start or pause a lab, reveal a hint, follow a path, draft and preview a Studio scenario, or read and export a candidate report. Billing, organization settings, invitations and your profile are not available through MCP, and you can revoke the connection from your profile at any time. The MCP documentation lists every tool.
Here is what that looks like in one of our own Claude Code sessions. We gave the agent a few lines of course notes and asked it, in French, to turn them into a graded Studio lab on a single-node kubeadm cluster, then preview the draft and test every check without publishing:

The agent created and validated the draft, launched a preview, and confirmed that all three tasks failed on the broken cluster. It then applied a reference solution to the preview, watched every task pass at 40 of 40, removed its temporary solution step, and left the draft unpublished for a human to review:

Plans and credits
A real lab burns real resources, so lab time is metered in credits. Labs draw 1 to 6 credits an hour depending on the environment, paused time is not charged, a failed provisioning is not billed, and no overage is ever invoiced. When credits run out, labs pause.
| Plan | Price | Credits | What it adds |
|---|---|---|---|
| Community | Free | 10, granted once | Tutorials, challenges and playgrounds on community scenarios |
| Pro | €25 per seat per month | 60 per seat per month | The full catalog for every seat holder |
| Instructor | €50 per seat per month | 100 per seat per month | Scenario Studio and instructor-led cohorts |
| Enterprise | By contract | Agreed in the contract | Proctored hiring assessments, SSO, private scenarios, every scenario for every member with no seats to assign |
Prices are those listed on klastro.io in November 2026; yearly billing gives two months free on Pro and Instructor.
How to get started
Klastro runs in production as a public preview, and accounts open by invitation, in waves, so every newcomer gets labs that start. Request an invitation with one line on what you want to do, then sign in at app.klastro.io. The user documentation covers every screen in this walkthrough in more detail.
Klastro is also where our own engineering shows. Its backend is developed with the local Kubernetes inner loop we described with Tilt, and its delivery follows the platform engineering practice we bring to clients. If you want your team trained on real clusters, or a hiring process that tests the work rather than the vocabulary, talk to us.
