Skip to content

ViDoRAG

A multimodal RAG framework for visually rich documents

Visit official site

What ViDoRAG is

The framework is described as a multimodal retrieval-augmented generation system for complex reasoning over visually rich documents, built by the Alibaba NLP team.

What you can do with it

  • Run multimodal retrieval over visual and text data
  • Apply a GMM-driven retrieval strategy
  • Support multi-hop reasoning with several agents

Getting started

  • Clone the project and install the dependencies
  • Build an index database with the ingestion script
  • Launch multi-agent inference and generation

Pros & cons

✓ What we like

  • Open-source code and datasets are offered
  • Retrievers and generators can be customised

! What to watch out for

  • The framework targets visually rich documents
  • Reasoning runs through multiple agents

FAQ

Who developed it?

The Alibaba NLP team is named.

What drives the retrieval strategy?

A Gaussian mixture model, or GMM, is described.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review