Module context_builder 

Memory-aware prompt construction engine (T8)

Assembles retrieved memories into structured LLM prompts with configurable token budgets, section priorities, and fill strategies.

Example

use engram::intelligence::context_builder::{
    ContextBuilder, PromptTemplate, Section, SimpleTokenCounter, Strategy,
};

let counter = SimpleTokenCounter;
let builder = ContextBuilder::new(Box::new(counter));

let template = PromptTemplate {
    sections: vec![
        Section { name: "System".into(), content: "You are helpful.".into(), max_tokens: 100, priority: 0 },
        Section { name: "Memories".into(), content: String::new(), max_tokens: 500, priority: 1 },
    ],
    total_budget: 600,
    separator: "\n\n---\n\n".into(),
};

let result = builder.build(&template, &memories, Strategy::Greedy);

Structs

ContextBuilder

Context builder that assembles memories into structured prompts.

MemoryEntry

Minimal memory representation used by the builder.

PromptTemplate

Template defining prompt structure.

Section

A section in the prompt with its own token budget and priority.

SimpleTokenCounter

Simple token estimator that assumes ~4 characters per token.

Enums

Strategy

Strategy for filling sections with memories.

Traits

TokenCounter

Abstraction for counting tokens in a string.