parthsarthi03/raptor

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1750 stars · 234 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

The official Python implementation of RAPTOR (Recursive Abstractive Processing for Tree-Organized Retrieval), a retrieval-augmented generation method from an arXiv paper. It recursively embeds, clusters, and summarizes document chunks into a multi-level tree, then retrieves from that tree at query time to answer questions with context spanning different levels of abstraction.

Use cases

• improve rag retrieval over long documents
• answer questions that require multi-step reasoning across a whole document
• build recursive hierarchical summaries of text for retrieval
• implement tree-organized retrieval in python
• get document-level context instead of short contiguous chunks
• question answering on benchmarks like QuALITY with GPT-4

When to choose

• You need retrieval that captures holistic document context rather than isolated chunks
• You are doing long-document or multi-hop question answering where standard RAG underperforms
• You want a programmable RAG layer where you can swap in custom summarization, embedding, or QA models

When to avoid

• You need a production RAG platform with ingestion pipelines, connectors, and UIs out of the box
• Your corpus is small or short documents where plain chunk retrieval already suffices and tree-building LLM costs are not justified
• You want a no-code turnkey search product rather than a library to integrate into your code

Facets

library · maturity active

rag machine-learning nlp search-engine artificial-intelligence machine-learning large-language-models python retrieval-augmented-generation tree-structured-retrieval recursive-summarization hierarchical-clustering question-answering long-document-retrieval research-paper-implementation document-indexing natural-language-processing search

Member repositories

For agents

markdown · JSON · MCP: product_card(name="parthsarthi03/raptor")