agentic_rag library

Retrieval-augmented generation for the agentic framework.

Documents in, cited answers out — with every stage of the pipeline replaceable.

import 'package:agentic_rag/agentic_rag.dart';

final indexer = RagIndexer(
  index: EmbeddingIndex(model: embeddings, store: store),
  chunker: const MarkdownChunker(),
);
await indexer.indexAll(documents);

final pipeline = RagPipeline(
  retriever: VectorRetriever(index: indexer.index),
  model: chatModel,
  reranker: const ScoreFloorReranker(minScore: 0.25),
);

final answer = await pipeline.answer('How do refunds work?');
print(answer.text);                 // "... refunds take 5 days [2]."
print(answer.citations.first.label); // "[2] Handbook › Refunds"

Each stage is a port: Chunker, Retriever, Reranker, DocumentLoader. Replacing one is a class in your own package, and nothing here needs to change.