Struct ContextWindowManager 

pub struct ContextWindowManager { /* private fields */ }

Context window management for RAG

Helps manage token limits when constructing prompts with retrieved documents.

Implementations

impl ContextWindowManager

pub fn new(max_tokens: usize) -> Self

Create new context window manager

Arguments

• max_tokens - Maximum tokens available for context (e.g., 4096 for GPT-3.5)

pub fn fit_documents<F>( &self, documents: Vec<Neighbor>, token_estimator: F, reserved_tokens: usize, ) -> Vec<Neighbor>

where F: Fn(&str) -> usize,

Fit documents into context window

Greedily adds documents until hitting token limit.

Arguments

• documents - Ranked documents to include
• token_estimator - Function to estimate tokens in text
• reserved_tokens - Tokens reserved for prompt template, query, etc.

Returns

Documents that fit within the context window

pub fn simple_token_estimator(text: &str) -> usize

Simple token estimator (words * 1.3)

Rough approximation. For production, use tiktoken or similar.