cupel

Context window management pipeline for LLM applications.

Given a set of context items (messages, documents, tool outputs, memory) and a token budget, Cupel determines the optimal context window — maximizing information density while respecting the attention mechanics of any LLM.

Features

• Fixed pipeline: Classify → Score → Deduplicate → Slice → Place
• Composable scoring: 6 built-in scorers + composite + scaled
• Multiple slicing strategies: Greedy, Knapsack, Quota, Stream
• Attention-aware placement: U-shaped and chronological placers
• Full explainability: Every inclusion/exclusion has a traceable reason
• Declarative policies: Named presets for common use cases

Usage

use cupel::prelude::*;

// Coming soon — source files migrate in Phase 17

License

MIT