Struct SnapcompactCompactor 

pub struct SnapcompactCompactor { /* private fields */ }

Bitmap-frame compactor that renders the discarded tail of a conversation as PNG frames via the snapcompact pipeline.

Unlike the LLM compactor, this compactor makes no LLM call — the local renderer is the only cost. Frames are addressed at vision-capable models (Anthropic, OpenAI, Google) that read the PNGs back directly.

Limits

Snapcompact always returns real PNG bytes for each frame or surfaces the failure via CompactionError::LlmError. There is no “no-op renderer” fallback: a rendering failure becomes a compaction failure (intentional, so the operator notices).

Implementations

impl SnapcompactCompactor

pub fn new() -> Self

Construct with snapcompact’s model-aware shape resolution.

pub fn with_shape(self, shape: Shape) -> Self

Pin a specific shape (overrides model-aware resolution).

pub fn with_frame_chars(self, frame_chars: usize) -> Self

Set the per-frame source-text character limit.

pub fn with_max_frames(self, max_frames: u32) -> Self

Set the maximum number of frames per compaction.

pub fn shape(&self) -> Option<&Shape>

Borrow the configured shape (if pinned).

Trait Implementations

impl Compactor for SnapcompactCompactor

fn estimate_tokens(&self, messages: &[Message]) -> usize

Estimate the token count of messages

fn compact<'a>( &'a self, messages: &'a [Message], instruction: Option<&'a str>, ) -> Pin<Box<dyn Future<Output = Result<CompactedContext, CompactionError>> + Send + 'a>>

Compact messages, returning a summary and kept messages

impl Default for SnapcompactCompactor

fn default() -> Self

Returns the “default value” for a type.