Struct ChunkingDistiller 

pub struct ChunkingDistiller<D: Distiller> { /* private fields */ }

A Distiller decorator that bounds the size of any single prompt sent to inner.

A long transcript is split into chunks of at most max_chunk_chars (via split_on_budget), each chunk is distilled independently through inner, and the partial results are combined into one DistilledMemory (via an internal reduce step, using the configured ReduceStrategy). A transcript that already fits in one chunk — including an empty transcript — is passed through to inner unchanged, with no splitting or reducing.

max_chunk_chars is caller-supplied policy: this type has no opinion on what a safe prompt size is for any particular model or host, only on how to split, map, and reduce once a budget is given.

Implementations

impl<D: Distiller> ChunkingDistiller<D>

pub fn new(inner: D, max_chunk_chars: usize) -> Self

Wraps inner, splitting any transcript over max_chunk_chars into multiple distillation calls. Uses ReduceStrategy::Structural to combine the results; call Self::with_reduce_strategy to use ReduceStrategy::Llm instead.

pub fn with_reduce_strategy(self, strategy: ReduceStrategy) -> Self

Sets the ReduceStrategy used to combine chunk results.

Trait Implementations

impl<D: Distiller> Distiller for ChunkingDistiller<D>

fn distill(&self, transcript: &str) -> Result<DistilledMemory>

Errors

Returns an error if any chunk’s call to inner.distill fails, or if ReduceStrategy::Llm is used and the consolidating call fails. On error, no partial result is produced: chunks already distilled successfully before the failure are discarded, not saved or returned.