Crate trueno_rag 

Trueno-RAG: Pure-Rust Retrieval-Augmented Generation Pipeline

This crate provides a complete RAG stack built on Trueno compute primitives with zero Python/C++ dependencies.

Quick Start

use trueno_rag::{
    pipeline::RagPipelineBuilder,
    chunk::RecursiveChunker,
    embed::MockEmbedder,
    rerank::NoOpReranker,
    fusion::FusionStrategy,
    Document,
};

// Build a RAG pipeline
let mut pipeline = RagPipelineBuilder::new()
    .chunker(RecursiveChunker::new(512, 50))
    .embedder(MockEmbedder::new(384))
    .reranker(NoOpReranker::new())
    .fusion(FusionStrategy::RRF { k: 60.0 })
    .build()
    .unwrap();

// Index a document
let doc = Document::new("Machine learning enables computers to learn from data.")
    .with_title("ML Intro");
pipeline.index_document(&doc).unwrap();

// Query the pipeline
let results = pipeline.query("machine learning", 5).unwrap();
assert!(!results.is_empty());

Chunking Strategies

Multiple chunking strategies are available:

• RecursiveChunker - Hierarchical splitting (default)
• FixedSizeChunker - Character-based splitting
• SentenceChunker - Sentence-boundary aware
• ParagraphChunker - Paragraph grouping
• SemanticChunker - Embedding similarity-based
• StructuralChunker - Header/section-aware

Fusion Strategies

Combine dense and sparse retrieval results:

• FusionStrategy::RRF - Reciprocal Rank Fusion (recommended)
• FusionStrategy::Linear - Weighted combination
• FusionStrategy::DBSF - Distribution-based score fusion

Example: Custom Chunking

use trueno_rag::{chunk::{ParagraphChunker, Chunker}, Document};

let chunker = ParagraphChunker::new(2); // 2 paragraphs per chunk
let doc = Document::new("Para 1.\n\nPara 2.\n\nPara 3.");
let chunks = chunker.chunk(&doc).unwrap();
assert_eq!(chunks.len(), 2);

Re-exports

pub use chunk::Chunk;

pub use chunk::ChunkId;

pub use chunk::ChunkMetadata;

pub use chunk::Chunker;

pub use chunk::ChunkingStrategy;

pub use chunk::FixedSizeChunker;

pub use chunk::ParagraphChunker;

pub use chunk::RecursiveChunker;

pub use chunk::SemanticChunker;

pub use chunk::SentenceChunker;

pub use chunk::StructuralChunker;

pub use embed::Embedder;

pub use embed::EmbeddingConfig;

pub use embed::PoolingStrategy;

pub use error::Error;

pub use error::Result;

pub use fusion::FusionStrategy;

pub use index::BM25Index;

pub use index::SparseIndex;

pub use index::VectorStore;

pub use metrics::AggregatedMetrics;

pub use metrics::RetrievalMetrics;

pub use pipeline::ContextAssembler;

pub use pipeline::RagPipeline;

pub use rerank::Reranker;

pub use retrieve::HybridRetriever;

pub use retrieve::RetrievalResult;

Modules

chunk

Document chunking strategies for RAG pipelines

embed

Embedding generation for RAG pipelines

error

Error types for Trueno-RAG

fusion

Score fusion strategies for hybrid retrieval

index

Indexing for RAG pipelines (BM25 sparse index and vector store)

metrics

Retrieval evaluation metrics

pipeline

RAG Pipeline implementation with context assembly

rerank

Reranking module for RAG pipelines

retrieve

Retrieval module for RAG pipelines

Structs

Document

A document to be indexed

DocumentId

Document identifier