Struct LlmCompressor 

pub struct LlmCompressor { /* private fields */ }

Summarizes old messages using an LLM provider.

Makes exactly one LLM call per compression event. If the call fails, falls back to rule-based pruning (remove oldest non-system messages).

Example

use traitclaw_core::context_managers::LlmCompressor;
let compressor = LlmCompressor::new(provider);

Implementations

impl LlmCompressor

pub fn new(provider: Arc<dyn Provider>) -> Self

Create a compressor with a given provider.

pub fn with_prompt(self, prompt: impl Into<String>) -> Self

Set a custom summarization prompt template.

pub fn with_threshold(self, threshold: f64) -> Self

Set the compression threshold (0.0–1.0).

pub fn with_keep_recent(self, count: usize) -> Self

Set the number of recent messages to keep verbatim.

Trait Implementations

impl ContextManager for LlmCompressor

fn prepare<'life0, 'life1, 'life2, 'async_trait>( &'life0 self, messages: &'life1 mut Vec<Message>, context_window: usize, state: &'life2 mut AgentState, ) -> Pin<Box<dyn Future<Output = ()> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait,

Prepare the message list by pruning or compressing if necessary.

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

Estimate the total token count for a message list.