Agent-level action RAG

They configure action-level RAG on the Agent class. They are not the chat-turn corpus.

Two corpora

Empty knowledge is a no-op. An all-default @Retrieval is also a no-op, so existing agents stay unchanged.

Declarative

@AgentSpec(
    description = "Answers from the product handbook",
    knowledge = {
        @KnowledgeSource(
            name = "handbook",
            type = KnowledgeSource.KnowledgeType.FILE,
            path = "knowledge/handbook")
    },
    retrieval = @Retrieval(
        strategy = Retrieval.Strategy.HYBRID,
        sources = "handbook",
        queryParam = "question",
        topK = 5)
)
public final class HandbookAgent extends Agent {
    @Override
    protected LLMClient getLLM() {
        return new AnthropicClient("claude-sonnet-4-20250514");
    }

    @Override
    protected List<Role> getRoles() {
        return List.of(/* action Role with a String query parameter */);
    }
}

AgentSpecExtractor reads those members during initialization:

• duplicate KnowledgeSource.name() values fail initialization
• retrieval.topK < 0 fails initialization
• Agent.getKnowledgeSources() / getRetrieval() (@since 0.13.0) now default to that extractor in TnsAI 0.14.0

A builder that already added sources or set a RetrievalConfig wins. ConfigurableAgent overrides the Agent defaults with the builder list.

Not this page

• Role- or method-level @Retrieval walkthrough — still RAG and the declarative tutorial
• Chat-turn @ChatKnowledge — still the chat-level path on the RAG index
• Maven Central 0.14.1 — these members are in the published JARs