Class RagStore

Namespace

Mythosia.AI.Rag

Assembly

Mythosia.AI.Rag.dll

A pre-built, shareable RAG index. Build once, share across multiple AIService instances.

public class RagStore

Inheritance

object

RagStore

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Extension Methods

RagDiagnosticsExtensions.Diagnose(RagStore)

Methods

BuildAsync(Action<RagBuilder>, Func<IReadOnlyList<VectorRecord>, Task>?, CancellationToken)

Builds a RagStore by loading, splitting, embedding, and indexing all configured documents. The resulting store can be shared across multiple AIService instances.

public static Task<RagStore> BuildAsync(Action<RagBuilder> configure, Func<IReadOnlyList<VectorRecord>, Task>? onDocumentEmbedded = null, CancellationToken cancellationToken = default)

Parameters

configure Action<RagBuilder>

Configuration delegate for the RAG builder.

onDocumentEmbedded Func<IReadOnlyList<VectorRecord>, Task>

Optional callback invoked after each document's embedding is complete. When provided, replaces the default UpsertBatchAsync call — the callback receives the generated VectorRecord list and decides how to persist them. When omitted, records are saved to the configured store automatically.

cancellationToken CancellationToken

Cancellation token.

Returns

Task<RagStore>

Examples

// Default: embed and save to store automatically
var ragStore = await RagStore.BuildAsync(config => config
    .AddDocuments("./knowledge-base/")
    .UseOpenAIEmbedding(apiKey)
);

// With callback: replace vectors atomically per document
var ragStore = await RagStore.BuildAsync(config => config
    .AddDocuments(loader, filePath)
    .UseEmbedding(embeddingProvider)
    .UseStore(vectorStore),
    onDocumentEmbedded: records =>
        vectorStore.ReplaceByFilterAsync(
            VectorFilter.ByMetadata("full_path", filePath), records),
);

QueryAsync(string, RagQueryOptions?, CancellationToken)

Processes a query through the RAG pipeline with per-request query overrides.

public Task<RagProcessedQuery> QueryAsync(string query, RagQueryOptions? options, CancellationToken cancellationToken = default)

Parameters

query string

options RagQueryOptions

cancellationToken CancellationToken

Returns

Task<RagProcessedQuery>

QueryAsync(string, IReadOnlyList<ConversationTurn>?, RagQueryOptions?, CancellationToken)

Processes a query through the RAG pipeline with automatic query rewriting and per-request query overrides.

public Task<RagProcessedQuery> QueryAsync(string query, IReadOnlyList<ConversationTurn>? conversationHistory, RagQueryOptions? options, CancellationToken cancellationToken = default)

Parameters

query string

conversationHistory IReadOnlyList<ConversationTurn>

options RagQueryOptions

cancellationToken CancellationToken

Returns

Task<RagProcessedQuery>

QueryAsync(string, IReadOnlyList<ConversationTurn>?, CancellationToken)

Processes a query through the RAG pipeline with automatic query rewriting for multi-turn conversations. If a Mythosia.AI.Rag.RagStore.QueryRewriter is configured and conversation history is provided, the query is rewritten into retrieval-ready form before search, and retrieval-oriented keywords may also be derived.

public Task<RagProcessedQuery> QueryAsync(string query, IReadOnlyList<ConversationTurn>? conversationHistory, CancellationToken cancellationToken = default)

Parameters

query string

The current user query.

conversationHistory IReadOnlyList<ConversationTurn>

Previous conversation turns for context, or null to skip rewriting.

cancellationToken CancellationToken

Cancellation token.

Returns

Task<RagProcessedQuery>

QueryAsync(string, CancellationToken)

Processes a query through the RAG pipeline: embed → search → build context. Returns the request message content and references without calling an LLM.

public Task<RagProcessedQuery> QueryAsync(string query, CancellationToken cancellationToken = default)

Parameters

query string

cancellationToken CancellationToken

Returns

Task<RagProcessedQuery>

SetQueryRewriter(IQueryRewriter?)

Sets or clears the query rewriter at runtime. Pass null to disable query rewriting and retrieval keyword derivation.

public void SetQueryRewriter(IQueryRewriter? queryRewriter)

Parameters

queryRewriter IQueryRewriter

UpdateOptions(Action<RagPipelineOptions>)

Updates pipeline options at runtime without rebuilding the index.

public bool UpdateOptions(Action<RagPipelineOptions> configure)

Parameters

configure Action<RagPipelineOptions>

Returns

bool

Examples

store.UpdateOptions(opt =>
{
    opt.TopK = 8;
    opt.MinScore = 0.4;
    opt.RetrievalMultiplier = 3;
    opt.PromptTemplate = "Based on:\n{context}\n\nQuestion: {question}";
});

UpdateRetrievalStrategy(bool, float)

Switches the retrieval strategy at runtime between vector-only and hybrid search without rebuilding the entire pipeline or re-indexing documents. Returns false if the underlying pipeline does not support runtime updates.

public bool UpdateRetrievalStrategy(bool useHybridSearch, float vectorWeight = 0.5)

Parameters

useHybridSearch bool

True to enable hybrid (vector + BM25) search, false for vector-only.

vectorWeight float

Weight for vector similarity [0, 1] when hybrid is enabled. 0.5 = equal weight.

Returns

bool