Module context_compression 

Context window compression and summarization utilities.

Provides strategies for keeping LLM conversation context within token budget limits by dropping, summarizing, or windowing older messages.

Structs

CompressionResult

Summary of what happened during a compression pass.

ContextBudget

Tracks token usage against a fixed budget.

ContextCompressor

Applies a CompressionStrategy to a slice of messages.

Message

A single message in a conversation context.

Enums

CompressionStrategy

Strategy used when compressing a context window.

Functions

estimate_tokens

Rough token estimate: word count × 1.3 + punctuation count.

importance_score

Compute an importance score for a message based on recency and role.