RagPipeline class

Ingests documents into a DocumentStore and answers questions from them.

final kb = await AsyncPhoenixCollection.open('kb', dimensions: 1024);
final rag = RagPipeline(
  store: kb,
  embedder: OpenAICompatibleEmbedder.voyage(
      apiKey: voyageKey, model: 'voyage-3.5', dimensions: 1024),
  chat: AnthropicChatModel(apiKey: anthropicKey),
);
await rag.ingest(RagDocument('handbook', handbookText));
final answer = await rag.ask('How many vacation days do I get?');
print(answer.text);
for (final c in answer.citations) print('[${c.number}] ${c.documentId}');

Retrieval is hybrid by default — embeddings and BM25 keywords fused with RRF — which is markedly more robust than either alone for names, codes and rare terms; MMR then removes near-duplicate chunks.

Constructors

RagPipeline({required DocumentStore store, required Embedder embedder, required ChatModel chat, TextChunker? chunker, int k = 6, bool hybrid = true, double? mmr = 0.7, String systemPrompt = defaultSystemPrompt, int maxContextChars = 24000})

Creates a pipeline. embedder must produce vectors of the store's dimensionality.

Properties

chat → ChatModel

Writes the answers.

final

chunker → TextChunker

Splits documents into chunks.

final

embedder → Embedder

Embeds chunks and questions.

final

hashCode → int

The hash code for this object.

no setter, inherited

hybrid → bool

Fuse keyword (BM25) retrieval with vector retrieval.

final

k → int

Chunks retrieved per question.

final

maxContextChars → int

Most source characters placed in one prompt; lower-ranked sources beyond it are left out whole rather than cut mid-text.

final

mmr → double?

MMR diversity for retrieval, or null for plain relevance.

final

runtimeType → Type

A representation of the runtime type of the object.

no setter, inherited

store → DocumentStore

Where chunks live.

final

systemPrompt → String

Instructions for the model.

final

Methods

ask(String question, {int? k, Filter? filter, List<ChatMessage> history = const [], int? maxTokens}) → Future<RagAnswer>

Answers question from the most relevant chunks.

askStream(String question, {int? k, Filter? filter, List<ChatMessage> history = const [], int? maxTokens}) → Future<RagStream>

Like ask, streaming the answer; the sources are known up front.

buildMessages(String question, List<RagSource> sources, {List<ChatMessage> history = const [], List<RagSource>? included}) → List<ChatMessage>

The prompt for question over sources, after history.

ingest(RagDocument document) → Future<int>

Ingests one document; returns its chunk count.

ingestAll(Iterable<RagDocument> documents) → Future<int>

Ingests documents (embedding all their chunks in batched calls) and returns the total chunk count. Re-ingesting a document replaces it: new chunks are written first, then stale ones removed, so the document never disappears in between.

noSuchMethod(Invocation invocation) → dynamic

Invoked when a nonexistent method or property is accessed.

inherited

remove(String documentId) → Future<int>

Removes a document's chunks; returns how many were removed.

retrieve(String question, {int? k, Filter? filter}) → Future<List<RagSource>>

The chunks most relevant to question.

toString() → String

A string representation of this object.

inherited

Operators

operator ==(Object other) → bool

The equality operator.

inherited

Constants

defaultSystemPrompt → const String

Default systemPrompt.