Struct ModelBackedCompactor 

pub struct ModelBackedCompactor {
    pub model: Arc<dyn Model>,
    pub tokens_per_char: f32,
    pub keep_recent: usize,
    pub summary_max_tokens: u32,
}

Compactor that calls an LLM for the inferential stages and falls back to DefaultCompactor’s structural strategies for the computational ones.

Typical wiring:

let cheap = OpenAiCompat::with_key(providers::DEEPSEEK, "deepseek-v4-flash", key);
let compactor = ModelBackedCompactor::new(Arc::new(cheap));
let loop_ = AgentLoop::new(main_model).with_compactor(Arc::new(compactor));

Fields

model: Arc<dyn Model>

tokens_per_char: f32

keep_recent: usize

Keep the most recent N turns intact during semantic compaction.

summary_max_tokens: u32

Hard cap on the summary length the model is asked to produce.

Implementations

impl ModelBackedCompactor

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

Trait Implementations

impl Compactor for ModelBackedCompactor

fn budget(&self, ctx: &Context) -> Budget

fn compact<'life0, 'life1, 'async_trait>( &'life0 self, stage: CompactionStage, ctx: &'life1 mut Context, ) -> Pin<Box<dyn Future<Output = Result<(), CompactError>> + Send + 'async_trait>>

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