LLM Context Forge (Node.js/TypeScript)

Production-grade infrastructure for managing, chunking, and evaluating LLM contexts with exact mathematical determinism.

llm-context-forge provides native tools for token counting (using identical backing algorithms to OpenAI via js-tiktoken), priority-based context window algorithms, cost evaluation pipelines, and extractive context compression.

🚀 Features

• Exact Tokenization Constraints: Maps OpenAI logic flawlessly, and uses production-heuristic bounds for Anthropic & Node environments.
• Dynamic Context Assembly: ContextWindow & ConversationManager safely inject CRITICAL inputs while smartly trimming LOW items before they hit API limits.
• Extraction Algorithms: Shrink redundant conversations mathematically using native TF-IDF string abstractions.
• Precise Pricing Evaluation: Convert raw inputs into concrete fractional cent dollar values deterministically.

📦 Installation

npm install llm-context-forge

📖 Usage

Quick Start

import { TokenCounter, DocumentChunker, ContextWindow } from "llm-context-forge";

// Precise Token Calculations
const counter = new TokenCounter("gpt-4o");
console.log(counter.count("Hello World!")); // Output: 3

// Semantic Chunking Limits
const chunker = new DocumentChunker();
const chunks = chunker.chunk("Massive String block...", { maxTokens: 50 });

// Smart Window Packing
const window = new ContextWindow();
window.addBlock("System instructions here", 0); // 0 = CRITICAL priority
console.log(window.assemble({ maxTokens: 4000 }));

For advanced documentation, refer to the Full API Reference.

🛠 Contributing

This package strictly enforces exact bounds coverage testing using jest. Tests map exactly to the mathematical equivalents bound in its Python sister library.

git clone https://github.com/your-username/llm-context-forge-js.git
npm install
npm test