RektRAG introduces a lightweight, tree-based RAG designed for high-precision retrieval with minimal token overhead. RektRAG utilizes Docling for structured parsing and introduces the TOON format to optimize context window usage.

By decoupling the core logic from specific LLM providers, RektRAG allows integration with any model. It leverages asynchronous processing and hierarchical summarization to provide a "No-brainer" experience for complex document retrieval.

Installation

Install using pip:

pip install rektrag

Usage

• RektEngine: Orchestrates document ingestion, state management, and multi-document retrieval.
• LLMProvider: Easily plug in OpenAI, Anthropic, or local LLMs by implementing the interface.

Example

Please refer to the Examples provided for further clarification.