Struct ContextualEnhancer 

pub struct ContextualEnhancer { /* private fields */ }

Contextual Retrieval enhancer (index-time transform).

L is any BaseChatModel; a cheap small model is recommended (one call per chunk). Enhancement failures are logged and degrade to the original content — see the module docs.

Implementations

impl ContextualEnhancer

pub fn new<L>(llm: L) -> Self

where L: BaseChatModel + Send + Sync + 'static, L::Error: Into<ProviderError>,

Creates an enhancer from any BaseChatModel.

pub fn new_arc( llm: Arc<dyn BaseChatModel<Error = ProviderError> + Send + Sync>, ) -> Self

Builds from an already-wrapped Arc<dyn BaseChatModel<Error = ProviderError>>.

pub fn with_config(self, config: ContextualConfig) -> Self

Sets the configuration.

pub fn with_max_concurrency(self, max_concurrency: usize) -> Self

Sets the max in-flight LLM calls.

pub async fn generate_context( &self, chunk: &str, ) -> Result<String, ContextualError>

Generates the situating context for one chunk (raw LLM access).

pub async fn enhance_document( &self, doc: &Document, ) -> Result<Document, ContextualError>

Enhances one document: returns a copy whose content is context + "\n" + original and whose metadata carries the context under CONTEXTUAL_METADATA_KEY.

Errors surface to the caller (use Self::enhance_documents for the fail-open batch path).

pub async fn enhance_documents(&self, docs: &[Document]) -> Vec<Document>

Batch enhancement with bounded concurrency and fail-open semantics.

Each chunk is enhanced concurrently (up to max_concurrency in flight); a chunk whose context generation fails keeps its original content (a warning is logged) — indexing never blocks on the enhancement. Idempotent: chunks already carrying CONTEXTUAL_METADATA_KEY are passed through untouched, so re-running over an indexed corpus is a no-op.

Trait Implementations

impl Debug for ContextualEnhancer

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.