Crate xz_rag 

Multi-channel Retrieval-Augmented Generation (RAG) engine.

xz-rag provides a composable RAG engine with multiple retrieval channels (semantic, BM25, metadata, graph), result fusion (RRF), context window management, query preprocessing (HYDE, expansion), and optional LLM generation.

Architecture

• Engine: DefaultRagEngine coordinating retrieval, fusion, and generation
• Channels: semantic, bm25, metadata, graph retrieval backends
• Pipeline: Channel orchestration with RRF fusion and score normalization
• Context: Token budget management and citation formatting
• Indexing: Document chunking strategies (fixed, recursive, separator-based)
• Caching: Optional in-memory result caching via moka
• Generation: LLM integration via xz-provider (feature-gated)

Re-exports

pub use engine::DefaultRagEngine;

pub use engine::DefaultRagEngineBuilder;

pub use error::RagError;

pub use traits::RagEngine;

pub use types::chunk::Chunk;

pub use types::chunk::ChunkMetadata;

pub use types::config::RagConfig;

pub use types::config::RagEngineInfo;

pub use types::rag::BuiltContext;

pub use types::rag::ChatMessage;

pub use types::rag::ChatRole;

pub use types::rag::Citation;

pub use types::rag::CitationFormat;

pub use types::rag::ContextConfig;

pub use types::rag::PromptTemplate;

pub use types::rag::RagGenerationConfig;

pub use types::rag::RagRequest;

pub use types::rag::RagRequestBuilder;

pub use types::rag::RagResponse;

pub use types::rag::RagStreamEvent;

pub use types::rag::RagTokenUsage;

pub use types::rag::RequestOptions;

pub use types::retrieval::ChannelStats;

pub use types::retrieval::QueryPreprocessing;

pub use types::retrieval::RetrieveRequest;

pub use types::retrieval::RetrieveRequestBuilder;

pub use types::retrieval::RetrieveResult;

pub use types::retrieval::RetrievedChunk;

pub use types::retrieval::StructuredFilter;

pub use pipeline::channel::ChannelConfig;

pub use pipeline::channel::ChannelPipeline;

pub use pipeline::channel::ChannelType;

pub use pipeline::fusion::RrfFusion;

pub use pipeline::normalize::MinMaxNormalizer;

pub use pipeline::normalize::ZScoreNormalizer;

pub use context::citation::format_citations_numeric;

pub use context::token_budget::ContextBuilder;

pub use indexing::chunker::ChunkStrategy;

pub use indexing::chunker::fixed::FixedSizeChunker;

pub use indexing::chunker::recursive::RecursiveCharacterChunker;

pub use indexing::chunker::semantic::SeparatorChunker;

pub use indexing::DocumentIndexer;

pub use indexing::IndexDocument;

pub use channels::graph::GraphChannelExecutor;

pub use channels::graph::KnowledgeGraphSearch;

pub use channels::metadata::MetadataChannelExecutor;

pub use channels::metadata::MetadataStore;

pub use channels::semantic::Embedder;

pub use channels::semantic::SemanticChannelExecutor;

pub use channels::semantic::SemanticSearch;

Modules

cache

In-memory result caching.

channels

Retrieval channel implementations (semantic, BM25, metadata, graph).

context

Context assembly: token budget management and citation formatting.

engine

Default RAG engine implementation.

error

RAG error types.

generation

LLM generation integration.

indexing

Document indexing and chunking strategies.

pipeline

Multi-channel pipeline orchestration.

preprocessing

Query preprocessing (HYDE, expansion).

traits

Core trait definitions.

types

Shared type definitions.