Guides · ConfigHub Workshop

New to cub? Install the cub CLI first. Public catalog packages pull and render anonymously, and you sign in only once a command saves or changes ConfigHub data.

Five cool demos

Helm ignores my valuesThe assistant wrote my Helm values. Did they do anything? Three of seven did nothing, and Helm never said so.
AI overwrites my fixesThe assistant rewrote my app. Are my fixes still there? Then ConfigHub carries the fixes through the rewrite.
My app does not fit my platformWill my app run on the platform I picked? Refused for two real reasons, fixed on both sides, then checked.
My chart behaves differently in GitOpsMy Redis has run fine for a month. Is it safe to hand to GitOps? Three problems, then a reviewed alternative from the Catalog.
My app installs but will not startEvery check passes. Should I install it? The image no longer exists, said before the install.

Start with what you use

Keep your tools. First inspect the configuration; choose delivery changes only when you are ready.

Bring an existing app

Start from the chart or delivery system you already use.

I use Argo CD. Preview your apps, import a parallel copy, then review handover and recovery.
I use Flux. Preview your repository and move delivery one cluster at a time.
I use Helm. Render your own chart and values, then check and compare a change before installing.
Other GitOps delivery paths. Review how rendered configuration reaches your controller.

Deploy something small first

Use a bounded example to learn the whole chain before bringing production configuration.

Deploy a four-object app. Relate source, Base, dev variant, Release, Argo CD Application, and ready workload.
Deploy a two-component stack. Check, import, place, release, and verify both components.

Build a platform or fleet for your apps

Choose the services your apps need, then review the configuration before anything runs.

Build a platform with Kubara. Generate platform add-ons, inspect them and add ConfigHub review when you are ready.
Manage a fleet with Sveltos. This Guide starts from an existing Sveltos management cluster and takes delivery into ConfigHub in stages.
Compose a stack with an app. Choose components and check that they fit together.
I use NVIDIA: inspect an AICR platform configuration, or prepare NIM model serving. Local inspection is separate from proving a GPU deployment.

Show me what is running · Let me browse configuration in a UI

Show me an example app to see what changes in everyday work.

Deep Dives

Follow a longer journey in stages, from inspection to governed delivery. Start by reading; each Guide explains when an account or cluster is needed. These links also stay in their original topics.

How the model works. Flatten it, put it in an OCI bundle with a plan if it needs one, and load it into ConfigHub. Three steps on one short page, the same for every format.
Can AI fix my fleet and prevent the next outage?. See recorded repairs, approvals and prevention on a disposable Sveltos fleet. Read the story without setup, inspect the evidence, or plan the full two-to-three-hour demonstration.

Continue with your own tools

Bring your Argo CD apps into ConfigHub. Preview your existing setup without an account, then save, connect and review a change while Argo CD keeps deploying.
Bring your Flux fleet into ConfigHub. Preview your existing setup without an account, then save, connect and review a change while Flux keeps deploying.
Bring Kubara's managed add-ons into ConfigHub. Manage your clusters' add-ons with Kubara, and approve every change to them in ConfigHub. See them offline in about 15 minutes, then govern them with cub kubara, or check them with an app as a Workshop stack.
Bring your Sveltos fleet into ConfigHub. Preview exported fleet configuration, review ownership and connect live profiles in stages.

Topic based walkthroughs

Deploy the GPU Operator from the Catalog. One Helm chart from the Catalog in five steps. Install it, load it into ConfigHub, upgrade it, and test it on a GPU node.
Deploy one thing, and look before you do. Pick one Catalog entry, read the exact objects it installs, then save it in ConfigHub and release it to a local cluster. One dated run shows the end state.
Walk the whole model once, with Redis. One chart, Redis, walked through the Catalog, flattening, OCI, variants, promotion, policy and stacks in one pass, marking which steps are free and local.
Try it: Redis in ten minutes. Render one public Redis catalog package and inspect its exact Kubernetes objects without contacting ConfigHub Server or Kubernetes.
From one chart to a governed fleet in ten minutes. In ten minutes you go from checking one chart to generating a governed fleet, checking a whole inference platform on the way, most of it free and copy-paste.
Use ConfigHub Workshop with your AI agent. Install the ConfigHub Workshop agent skill, choose a configuration task, and keep exact objects, lifecycle work, checks, and limits visible.
Find a starting configuration. Say what you need, try a simple Redis example, choose how to run a starting example, then continue in ConfigHub or on a platform.
Check a claim yourself. Pick a claim the site makes, run the smallest check that answers it, and read what it proved and what it left open.
Inspect and keep an exact record. Download the full Redis default record with its Catalog and record hashes, then reproduce the lookup and a refusal locally with Node and a checkout.
Choose a tool and start. Pick the tool for what you want to do now, and the path for a chart the Catalog does not have.
See what is running with cub scout. Inspect one workload and its owner in a read-only cluster map, without a ConfigHub account.

Helm questions

Answer the ten questions Helm users ask. Walk the ten questions a Helm user asks, with the pain beneath each one and a check you run on your agent's answer.
Find why a Helm value did not change the output. Compare a typo with the correct key in controlled renders, including a value the template transforms.
Why did Helm ignore my values?. Find values that Helm accepts but a chart does not use by comparing the rendered Kubernetes objects with and without each supplied key.
Did this chart version change upstream?. Check whether an upstream publisher changed the package bytes behind an existing Helm chart version.
Did your Bitnami chart stop pulling?. Find a tested, verified successor for a Bitnami chart that no longer pulls anonymously, with each successor linked to its catalog entry.
Review a chart upgrade before promoting it. Compare retained versions and prepare a review packet that keeps missing promotion evidence visible.
See what a gpu-operator upgrade changes. Render three versions of NVIDIA's public gpu-operator chart, then compare a version upgrade, a patch upgrade and a driver version change field by field.
Adapt a configuration and inspect the exact edit. Change one replica count, inspect the exact diff and spot an unexpected second edit.
Render your own chart to exact objects, adapt them as data, and keep them. Render your own chart and values, an AI-written one included, then check, diff, adapt a field, and keep the reviewed objects for your own Argo CD or Flux.
Send a missing or broken public chart. Look for a public chart in the Catalog, render it yourself, and send the chart and values when it is missing or its output differs from Helm.

Formats

Answer the questions AICR users ask. Compose an H100 training recipe's exact objects, compare three retained versions, and read the recorded accelerator-fit and component-order routes.
Answer the questions Timoni users ask. Build a typed module's exact objects, check the typed schema and the CRD ordering, and see which module bytes are digest-bound.
Answer the questions plain Kubernetes YAML users ask. Check what four files will install, compare two versions with one edit, and read why this base's lifecycle-route status is a recorded gap.
Answer the questions Kubara users ask. Check what a generated Kubara platform contains, and read its recorded CRD, secret and bootstrap routes. To govern a Kubara platform, use cub kubara.

GitOps

Use ConfigHub With Argo CD Or Flux. Keep the Argo CD or Flux you already run. ConfigHub publishes the reviewed Kubernetes objects as an OCI image for that controller to pull.
Run it with Flux, Argo CD, or kubectl. Reconcile a published component with no account, render any catalog chart to an image your Flux, Argo CD, or kubectl consumes, and verify its receipt first.
Harden Argo CD before production, as a variant you can check and keep. Turn security advice into a checked variant of the Catalog's Argo CD base, compare it with the base, and keep it with a staging and production path.
Bring your Sveltos fleet into ConfigHub. Preview exported fleet configuration, review ownership and connect live profiles in stages.
Bring your Argo CD apps into ConfigHub. Preview your existing setup without an account, then save, connect and review a change while Argo CD keeps deploying.
Bring your Flux fleet into ConfigHub. Preview your existing setup without an account, then save, connect and review a change while Flux keeps deploying.

Stacks and platforms

Compose and review a local workshop stack. Save a Kubara and Argo CD selection, move it, edit the app and retain a refusal when an API does not fit.
Make a stack from Catalog entries. Compose NVIDIA's gpu-operator, nvsentinel and cluster-readiness-engine entries into one local stack, check them together, then swap one entry and check again.
Find the hook and CRD work before delivery. Inspect a hook without running it, preserve an API refusal and identify what delivery still requires.
Match a GPU workload with supplied facts. Keep candidate, mismatch and unknown results separately, with the input hashes behind each answer.
Compose and check a stack. Get a stack, check it for conflicts and missing needs before anything runs, then run and govern it in ConfigHub. It includes the stacks made to be refused.
Try NVIDIA AICR for GPUs. Compare existing GPU nodes with a recipe, or pull and read a retained AICR configuration with ORAS.
Put an app on a platform. Check what an app needs, check it on a platform, take it into ConfigHub, or bring an app that already runs.
Bring Kubara's managed add-ons into ConfigHub. Manage your clusters' add-ons with Kubara, and approve every change to them in ConfigHub. See them offline in about 15 minutes, then govern them with cub kubara, or check them with an app as a Workshop stack.
Handle hooks and CRDs. Find the hooks, CRDs, setup jobs and target facts a chart needs before delivery, decide who runs each, and stage the target first.

ConfigHub

Can I promote this configuration?. Compare current and proposed Kubernetes objects, see what changes, and choose the tests required before moving the change.
Review one change before it ships. List the objects one change affects in each environment with no account, then propose it in ConfigHub, see the gate refuse it, approve it, release it and read the record.
Add one field and keep the original configuration. Review one added label, detect an object replacement and restore the exact original file.
Does the cluster match what we approved?. Compare approved configuration with live cluster state while keeping the current field-coverage limits visible.
Why do development and production differ?. Record development and production as related configurations so their exact differences and promotion history remain visible.
Take a config further. Take a reviewed configuration into a stack, keep it as an image, or save it in ConfigHub to move it through environments.
Change a config safely. Decide whether a change rebuilds the base or belongs to one environment, then preview it and promote only what was reviewed.
Operate a fleet. Check the starting point, choose an operation, keep a fleet record, and open the working App demonstrations.
Build a ConfigHub App. Build an App that repeats one operation on configuration saved in ConfigHub.
Release, promote and roll back. Release a reviewed configuration, promote it to the next environment, gate it with approvals, and roll it back.
View and change configuration with ConfigHub UIs. Browse stored configuration in the browser or cub commander, compare environments and separate editing from release.

Need help choosing? Follow the configuration lifecycle

Find a Guide by the step you are on

Start with a configuration

Learn by doing

Three guides, shortest first. Each one runs real commands against real packages, so you finish with output you can check.

The object digest, the OCI manifest digest, and the release digest are three different records. See the digest glossary.

Run the short example: pull one public package, render it locally, and read the exact Kubernetes objects it produces, all on your machine. Work through an example like yours: start from your own Helm values, an AICR recipe, an OCI package, or Kubernetes YAML. Follow one package end to end: take a single Redis configuration through pulling, verifying, changing, and upgrading it.

First choose the answer you need

The same questions apply to every input format, from a Helm chart to plain Kubernetes YAML. Choose the question you need before choosing a command.

Question	What you need	What it does not prove
What do I have?	The source, snapshot, package, or exact files.	What a source tool will generate, or whether a destination can run it.
What will it produce?	The source inputs and the exact materialization command, or literal objects where materialization is a no-op.	Whether a named destination has the required APIs, credentials, hardware, and lifecycle setup.
Can this destination accept it?	The exact candidate and current facts from the named destination.	That the configuration has been deployed or works at runtime.
Did it work?	The exact delivered revision and the live result required by the claim.	Results for another revision, destination, or untested behavior.

A missing prerequisite marks the unanswered stage blocked or not run, which is a gap in coverage rather than a failure of the source or its result.

Can I try one simple package?

Try Redis for a short local exercise with no server, cluster, or account.

Can I try one AICR configuration without a GPU?

Pull and verify the seven-Application CPU starter, then write it as a local OCI without ConfigHub or Kubernetes.

Which public configuration should I use?

Use the Component Catalog to choose a component and exact retained package version, then inspect its packaged configurations, required setup, and evidence.

How do I bring my own input?

Worked Examples covers your own Helm values, an AICR recipe, or plain Kubernetes YAML.

How do I choose the right command?

See what cub helm template, cub helm install, cub installer, and the ConfigHub upload and variant commands each prove.

I know Helm. How do these commands map to it?

Read the cheat sheet from Helm habits such as --set and release names to cub's declared-input model.

Do I need a ConfigHub Server or account to start?

Pull a catalog package, render it locally, and keep the files under your own control with no server, account, or cluster.

How do I check my own Helm values or a result I do not understand?

Build a local prompt for the AI assistant you already use. Private charts and values stay on your machine.

How do I add ConfigHub to an existing Kubara platform?

Keep Kubara's component selection and generated topology, with Argo reconciliation intact, while following one six-step adoption tutorial with explicit evidence checkpoints.

Prepare it for deployment

Where should the files live?

Deployment explains the local files, OCI, and ConfigHub choices in one sequence.

How do I turn reviewed files into a deployable OCI?

Choose the local OCI or ConfigHub release path, then let Argo CD or Flux pull the reviewed objects.

What happens to hooks and CRDs?

See how required setup is recorded and ordered, then tested or marked blocked.

Should I change the source input or the rendered object?

Use this guide to decide whether a change belongs in Helm values or in the stored Kubernetes objects.

How do Argo CD and Flux receive it?

The GitOps guide explains how controllers pull reviewed Kubernetes objects from Git or OCI.

Change or operate saved configuration

How do I make environment variants?

Create, compare, and promote development and production variants.

How do I test a change before promotion?

Compare the exact current result with the candidate for the next stage. The review shows what changed, what blocks the move, and what still needs a staging test.

How do I upgrade and roll back?

The day-2 upgrade story: diff the value model first, check control points and immutable fields, then upgrade rendered bundles by digest. The Redis walkthrough shows one full upgrade, promotion, and rollback.

What can a ConfigHub App automate?

Apps on ConfigHub includes upgrade, RBAC, and fleet examples, among others.

How do I roll a change through a fleet?

Open the Kubara and Sveltos examples for platform configuration, cluster assignments, and rollout evidence.

How do I start from an existing application?

Start read-only from GitOps, Helm, or a live cluster.

What if an upstream registry or its terms change?

Repoint image references across environments with the digest intact, promote the change environment by environment, and prove where it landed.