Class DefaultChunkerConfig

java.lang.Object
ai.pipestream.module.pipelineprobe.pipelinecrawl.DefaultChunkerConfig
@ApplicationScoped public class DefaultChunkerConfig extends Object
Builds the chunker module's custom_config for a pipeline-crawl run. Lifted out of PipelineCrawlerImpl so the orchestrator doesn't carry hand-rolled Struct/Value building.

Today this exposes one preset: a sentence-10-3 chunker on the body field, with embedder_configs populated only when the embedder step is in the pipeline. When the harness adds a "Run with N chunker strategies" UI option, this bean grows additional builders; the orchestrator picks one by name.

Built using config_id + config pairs under a chunker_configs list, scoped to a body directive. The shape mirrors what the chunker module reads from vector_set_directives.

  • Constructor Details

    • DefaultChunkerConfig

      public DefaultChunkerConfig()
      Creates the default chunker config builder. Instances are managed by CDI.

  • Method Details

    • sentence10x3OnBody

      public com.google.protobuf.Struct sentence10x3OnBody(boolean withEmbedder)
      Sentence chunker, 10 sentences per chunk, 3-sentence overlap, body-only.
      Parameters:
      withEmbedder - when true, the body directive carries an embedder_configs list with the minilm default config so the embedder accepts SPRs from this directive. When false, the field is omitted entirely — the chunker-only path doesn't need it, and an empty list trips the embedder's §21 "every directive MUST carry at least one NamedEmbedderConfig" check.
      Returns:
      the chunker custom_config struct wrapping the body vector_set_directives