Crate vil_context_optimizer 

VIL Context Optimizer

Intelligently compresses and optimizes LLM context windows to fit more information in fewer tokens. Supports deduplication, importance scoring, and budget-aware chunk selection.

Example

use vil_context_optimizer::{
    ContextOptimizer, TokenBudget, OptimizeStrategy,
};

let budget = TokenBudget::new(8000)
    .system_tokens(500)
    .response_tokens(1000);

let optimizer = ContextOptimizer::new(budget)
    .strategy(OptimizeStrategy::Full { dedup_threshold: 0.8 });

let chunks = vec![
    ("Rust is a systems language.".into(), 0.9),
    ("Rust is a systems language.".into(), 0.9),  // duplicate
    ("Python is great for ML.".into(), 0.7),
];

let result = optimizer.optimize(&chunks);
println!("Kept {}/{} chunks, saved {} tokens",
    result.final_count, result.original_count, result.tokens_saved);

Re-exports

pub use budget::TokenBudget;

pub use dedup::deduplicate;

pub use optimizer::ContextOptimizer;

pub use optimizer::OptimizedContext;

pub use scorer::score_chunks;

pub use scorer::ChunkScore;

pub use scorer::ScoringWeights;

pub use strategy::OptimizeStrategy;

pub use plugin::OptimizerPlugin;

pub use semantic::OptimizeEvent;

pub use semantic::OptimizeFault;

pub use semantic::OptimizeFaultType;

pub use semantic::OptimizerState;

Modules

budget

Token budget management.

dedup

Content deduplication based on word-level Jaccard similarity.

handlers

VIL pattern HTTP handlers for the Context Optimizer plugin.

optimizer

Main context optimizer.

pipeline_sse

SSE pipeline builders for Context Optimizer operations.

plugin

VilPlugin implementation for Context Optimizer integration.

scorer

Chunk importance scoring.

semantic

Semantic types for Context Optimizer operations.

strategy

Optimization strategies for context window compression.