v0.3.1 - First Official Release 🎉

🎉 FractoP v0.3.1 - First Official Release

FractoP (Fractal Processor) is a production-ready library for processing infinite-length text with LLMs using fractal architecture.

✨ Key Features

• UnillM Integration: Seamless integration with @aid-on/unillm for unified LLM access
• Infinite Length Support: Process documents of any size through intelligent chunking
• Type-Safe: Full TypeScript support without using any types
• Enterprise-Grade: Built-in timeouts, retries, and circuit breaker pattern
• Flexible: Support for any LLM provider through adapters

📦 Installation

npm install @aid-on/fractop

🚀 Quick Start

With UnillM (Recommended)

import { createWithUnillM } from '@aid-on/fractop';
import { generate } from '@aid-on/unillm';

const processor = createWithUnillM(
  async (chunk) => {
    const result = await generate(
      'groq:llama-3.1-8b-instant',
      [
        { role: 'system', content: 'Summarize this text.' },
        { role: 'user', content: chunk }
      ],
      { groqApiKey: process.env.GROQ_API_KEY }
    );
    return result.text;
  },
  { chunkSize: 3000, overlapSize: 300 }
);

const summary = await processor.process(longText);

With Custom LLM

import { createLLMAdapter, FractalProcessor } from '@aid-on/fractop';

const adapter = createLLMAdapter(async (text) => {
  return await yourLLM.complete(text);
});

const processor = new FractalProcessor(adapter, {
  chunkSize: 3000,
  overlapSize: 300
});

📚 API Reference

• createWithUnillM() - Create processor optimized for UnillM
• createLLMAdapter() - Convert async functions to LLMProvider
• createUnillMProcessor() - Advanced metadata-based processor
• FractalProcessor - Core processor class
• simpleMerge() - Simple result merging utility
• weightedMerge() - Weighted result merging

🧪 Tested With

• GROQ API (Llama 3.1)
• Full TypeScript type checking
• Comprehensive test coverage
• Real-world usage scenarios

📖 Documentation

See the README for complete documentation and examples.

🙏 Credits

Built with ❤️ by the Aid-On Team