Struct ContextWindow 

pub struct ContextWindow {
    pub max_tokens: usize,
    pub system_tokens: Vec<u32>,
    pub conversation: Vec<u32>,
    pub strategy: TruncationStrategy,
}

A fixed-capacity token window with configurable truncation.

The window stores a protected system prompt (never truncated) and a mutable conversation segment. Together they must fit within max_tokens.

Fields

max_tokens: usize

Maximum total token count (system + conversation).

system_tokens: Vec<u32>

Tokens belonging to the system prompt — never truncated.

conversation: Vec<u32>

Accumulated conversation tokens.

strategy: TruncationStrategy

How to truncate when the window is full.

Implementations

impl ContextWindow

pub fn new(max_tokens: usize, strategy: TruncationStrategy) -> Self

Create a new empty context window.

pub fn set_system_prompt( &mut self, tokens: Vec<u32>, ) -> Result<(), ContextError>

Set the system prompt tokens.

Returns an error if the system prompt alone exceeds max_tokens.

pub fn append(&mut self, tokens: &[u32]) -> usize

Append tokens to the conversation.

If the new tokens cause the window to overflow, truncation is applied before appending as much of the new tokens as will fit.

Returns the number of tokens actually appended.

pub fn truncate_to_fit(&mut self) -> usize

Truncate the conversation segment to make the total fit within max_tokens.

Applies the configured TruncationStrategy. Returns the number of tokens removed from the conversation.

pub fn tokens(&self) -> Vec<u32>

Concatenate system tokens and conversation tokens into a single flat vector.

The result is always within max_tokens.

pub fn len(&self) -> usize

Total token count (system + conversation).

pub fn is_empty(&self) -> bool

Returns true if both system and conversation are empty.

pub fn remaining_capacity(&self) -> usize

Number of additional tokens that can be appended before truncation.

pub fn is_at_limit(&self) -> bool

Returns true if the window is at or beyond its maximum capacity.

pub fn clear_conversation(&mut self)

Clear all conversation tokens (system prompt is preserved).

pub fn utilization(&self) -> f32

Fraction of max_tokens currently in use: len / max_tokens.

Returns 0.0 if max_tokens is zero.