OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
RAG & Knowledge Systems

Visrag

Visrag is an open-source Python framework for visual retrieval-augmented generation. It embeds document pages as images with vision-language models, retrieves relevant images, and uses them as context for generation. Its EVisRAG workflow adds evidence-guided reasoning across multiple images.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Visrag — OpenViST project profile in RAG & Knowledge Systems.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Visrag when a document question-answering workflow should retrieve pages as images rather than rely on text parsed from documents, or when a question requires evidence from multiple images.

Who it is for

  • Developers building retrieval systems for document images or PDF pages.
  • Researchers exploring multimodal retrieval and evidence-guided question answering.

Documented uses

  • Retrieve relevant document images to provide context for a vision-language model.
  • Prototype question answering over collections of visual documents.
  • Explore multi-image workflows that record per-image evidence before producing an answer.
Before you choose

Requirements & limits

The project is available as a public GitHub repository and identifies its license as Apache-2.0. The README links to a hosted pipeline and a Google Colab notebook.

  • The standard VisRAG setup specifies Python 3.10.8 and CUDA Toolkit 11.8; EVisRAG has a separate setup using Python 3.10.
  • VisRAG-Ret is built on MiniCPM-V 2.0; the README says the generator can be selected from vision-language models and names MiniCPM-V 2.0, MiniCPM-V 2.6 and GPT-4o as examples.
  • EVisRAG is described for multi-image questions and records image-specific evidence before reasoning.
  • The repository links to model and benchmark resources on Hugging Face, as well as a hosted pipeline and Colab notebook.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Prepare the VisRAG environment

Install the repository’s documented VisRAG dependencies and editable packages.

Before you start

  • Git
  • Conda
  • Python 3.10.8
  • CUDA Toolkit 11.8

Steps

  1. Clone the OpenBMB/VisRAG repository.

  2. Create and activate a Conda environment named VisRAG using Python 3.10.8.

  3. Install CUDA Toolkit 11.8 through Conda, then install the repository requirements and VisRAG in editable mode.

  4. Install the included timm_modified package in editable mode.

Expected outcome

A Conda environment with the documented VisRAG dependencies and local packages installed.

Guide limitations

  • This covers setup only; it does not configure a corpus, run retrieval or verify hardware compatibility.
  • The README describes timm_modified as supporting gradient checkpointing used in training to reduce memory use.
Documentation 1 ↗
COMPARE OPTIONS

Compare options for Retrieval-augmented applications

These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.

ProjectDocumented purposePlatformAccess and requirements
VisragRetrieve relevant document images to provide context for a vision-language model.The project is available as a public GitHub repository and identifies its license as Apache-2.0. The README links to a hosted pipeline and a Google Colab notebook.
LangsearchCrawl website or local data for a knowledge base.CLI, Self-hosted, Server
Hybrid Search RagCombine document retrieval and language-model generation in a configurable workflow.
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
9 October 2026
Original website title
OpenBMB/VisRAG · Official homepage title · retained in its source language
How we review sourcesSuggest a correction