Struct LlmCompactor 

pub struct LlmCompactor { /* private fields */ }

Compaction strategies and managers for long conversations. LLM-based compactor that uses the model itself to summarize

Implementations

impl LlmCompactor

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

Create a new LLM compactor with default configuration

pub fn with_config( model: Model, provider: Arc<dyn Provider>, config: CompactionConfig, ) -> Self

Create a new LLM compactor with custom configuration

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

Set how many recent messages to always keep

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

Set maximum batch size for summarization

pub fn with_target_ratio(self, ratio: f32) -> Self

Set target compaction ratio

impl LlmCompactor

Additional methods for LlmCompactor (not part of Compactor trait)

pub async fn summarize_branch( &self, messages: &[Message], branch_name: &str, ) -> Result<String, CompactionError>

Summarize a conversation branch for comparison purposes.

This is used when branching occurs and you want to understand what changed compared to another branch (e.g., main).

Trait Implementations

impl Compactor for LlmCompactor

fn compact<'life0, 'life1, 'life2, 'async_trait>( &'life0 self, messages: &'life1 [Message], instruction: Option<&'life2 str>, ) -> Pin<Box<dyn Future<Output = Result<CompactedContext, CompactionError>> + Send + 'async_trait>>

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

Compact messages, returning a summary and kept messages

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

Estimate the token count of messages