Struct ContextCompressor 

pub struct ContextCompressor { /* private fields */ }

Adaptively compresses a set of memories to fit within a token budget.

Implementations

impl ContextCompressor

pub fn new(budget_tokens: usize) -> Self

Create a new compressor with the given token budget.

pub fn estimate_tokens(text: &str) -> usize

Estimate the token count for text using the heuristic chars / 4.

pub fn compress_light(text: &str) -> String

Light compression: remove filler phrases (case-insensitive), then drop standalone stopwords and collapse extra whitespace.

pub fn compress_medium(text: &str) -> String

Medium compression: keep the first sentence of each paragraph and any sentence that contains a word starting with a capital letter (entity heuristic).

pub fn compress_heavy(text: &str) -> String

Heavy compression: extract lines that look like key facts — “Entity: something”, lines containing numbers, or dates.

pub fn compress_single(content: &str, level: CompressionLevel) -> String

Apply the specified compression level to content.

pub fn compress_for_context( memories: &[MemoryInput], budget: usize, ) -> Vec<CompressedEntry>

Adaptively compress a slice of memories to fit within budget tokens.

Algorithm:

1. Sort memories by importance (descending) — most important get processed first and receive lighter compression.
2. For each memory, try compression levels in order: None → Light → Medium → Heavy.
3. The first level that fits the remaining budget is used.
4. If even Heavy doesn’t fit, the memory is skipped.

Returns the ordered list of successfully compressed entries.

pub fn budget(&self) -> TokenBudget

Current budget state of this compressor instance.

Note: compress_for_context is a free function that does not mutate the compressor. Call this after manually tracking usage with estimate_tokens, or use it to inspect the configured budget.