Spring AI Add a MustacheTemplateRenderer (with optional constraint hook) for prompt templating

Please do a quick search on GitHub issues first, the feature you are about to request might have already been requested.

Expected Behavior

Spring AI should expose a TemplateRenderer backed by the Mustache JVM implementation so prompts can leverage loops, dotted notation, and section/conditional blocks.

Typical usage

val renderer = MustacheTemplateRenderer.builder().build()

val prompt = renderer.apply(
    """
    Query: {{query}}

    {{#items}}
    - {{name}} ({{pricing.salePrice}})
    {{/items}}
    """.trimIndent(),
    mapOf("query" to userQuery, "items" to rerankedItems)
)

Internally the renderer could follow this structure

class MustacheTemplateRenderer(
    private val constraints: List<Constraint> = emptyList(),
    private val mustacheFactory: DefaultMustacheFactory = DefaultMustacheFactory()
) : TemplateRenderer {

    override fun apply(template: String, variables: Map<String, Any?>): String {
        val sanitized = constraints.fold(variables, ::applyConstraint) // optional hook for apps
        val mustache = mustacheFactory.compile(StringReader(template), "prompt")
        return StringWriter().also { mustache.execute(it, sanitized) }.toString()
    }
}

That constraints.fold hook isn’t required for a first release, but offering some extension point (or builder slot) would let applications plug in limit/filter/range-style processors before rendering without forking the renderer.

Current Behavior

The only built-in renderer today is the StringTemplate (ST) version (spring-ai-template-st/src/main/java/org/springframework/ai/template/st/StTemplateRenderer.java:32).

ST excels at simple key replacement but doesn’t provide Mustache semantics like section blocks ({{#items}}…{{/items}}), inverted sections ({{^items}}), or dotted access ({{item.price.value}}), so expressing structured prompt layouts is cumbersome.

There’s also no standard way to preprocess collection variables before rendering.

Context

Several teams need Mustache-style templating to keep prompts readable, iterate through structured data, and optionally filter/limit lists before they reach the LLM.

A first-party MustacheTemplateRenderer (similar to the ST module) would remove the need for each project to reimplement it and would let us enable it via configuration.

Comment From: ghostg00

+1