YAML and JSON hold your data, and templating tools produce more of it, but nothing checks the result. CUE does not replace them. It adds the rules your files must follow and checks every file against those rules, so mistakes show up before deployment, not in production.
A data templating language that builds JSON by extending and overriding objects. It is dynamically typed, with no schemas; checks are hand-written assertions.
In CUE, types and rules sit beside the values they govern, and rules from different teams combine instead of overriding each other.
A typed configuration language with its own syntax. Modules build on each other by inheritance and amending, where later values override earlier ones.
CUE has no inheritance: rules from any number of sources combine in any order. Valid JSON is already valid CUE, so you start from your files.
Go text templates that paste values into YAML. A chart’s optional values schema checks its inputs, not the manifests its templates produce.
CUE works on structure, not text, and refuses to output configuration that breaks its rules. It can check Helm values and rendered manifests too.
Patches YAML in layers: a base plus an overlay per environment, applied in order, with later patches winning. Your own rules need a separate checker.
In CUE, layers can’t silently override each other. They all hold at once, and a conflict is an error, not a surprise in production.
A separate policy language, typically used to check configuration after another tool has produced it.
CUE defines, generates, and checks configuration in one language. The same rules shape your data and check it, and they read like the data they constrain.
A vocabulary for describing and validating JSON data, itself written in JSON or YAML.
CUE expresses the same constraints in a fraction of the lines: quicker for people to read, and fewer tokens for an AI to read or write.
Check existing files without rewriting them, and move to CUE one file at a time.
Write your organization’s policies as CUE rules: required fields, allowed values, limits, and naming. Every change is checked against them, on a laptop or in CI, and violations are reported before anything ships.
CUE and AI
Agents now write infrastructure, Kubernetes manifests, and CI pipelines. Their output looks plausible but is never 100% reliable, and there is more of it than people can review.
A second model can check the first, but its verdict is a probability. CUE’s is not: the same rules give the same answer, every time, and when a rule fails it says which one.
A wrong value often passes tests and only fails in production, under load or in one region. CUE checks it the moment the change is written.
A CUE schema tells an agent what valid looks like. It runs cue vet, sees what broke, and fixes it before a person reviews the change.
Step 1
Homebrew, a binary release, a container image, or a build from source. The install page lists every way to get the cue command.
Step 2
Say what a valid service looks like.
#Service: {
name: string
replicas: int & >0 & <=10
}Step 3
CUE reports every value that breaks the rule.
cue vet -d '#Service' \
schema.cue service.yamlInstall CUE, or talk to us about CUE at your company.