Struct SlidingWindowTransformer 

pub struct SlidingWindowTransformer { /* private fields */ }

Built-in sliding window context transformer with compaction reporting.

Wraps the same logic as sliding_window but captures compaction metrics for reporting.

Accepts an optional TokenCounter for pluggable token estimation. When none is provided, the default chars / 4 heuristic is used.

Implementations

impl SlidingWindowTransformer

pub fn new(normal_budget: usize, overflow_budget: usize, anchor: usize) -> Self

Create a new sliding window transformer.

Arguments

• normal_budget - Token budget under normal operation.
• overflow_budget - Smaller token budget used when overflow is signaled.
• anchor - Number of messages at the start to always preserve.

pub fn with_token_counter(self, counter: Arc<dyn TokenCounter>) -> Self

pub const fn with_cached_prefix_len(self, len: usize) -> Self

Set the cached prefix length to protect from compaction.

When caching is active, the effective anchor is max(anchor, cached_prefix_len).

pub const fn set_cached_prefix_len(&mut self, len: usize)

Update the cached prefix length (for runtime updates from the turn pipeline).

Trait Implementations

impl ContextTransformer for SlidingWindowTransformer

fn transform( &self, messages: &mut Vec<AgentMessage>, overflow: bool, ) -> Option<CompactionReport>

Transform the context messages in-place.