Repolode
Submit
NVIDIA

NVIDIA/NeMo-Retriever

A scalable pipeline that extracts text, tables, charts and images from documents into structured JSON, using NVIDIA NIM microservices and models, for RAG applications.

View on GitHubWebsite
Stars
2,988
Forks
357
Open issues
258
Last commit
2 days ago

What is it?

NeMo Retriever Library is a document content and metadata extraction framework from NVIDIA. It splits documents into pages, classifies artifacts such as text, tables, charts and infographics, and runs OCR to produce a defined JSON schema. The pipeline can then compute embeddings for the extracted content and store them in LanceDB, preparing data for retrieval-augmented generation.

The pipeline can run through NVIDIA NIM microservices or a range of local models. A small-scale Python library mode works with Hugging Face models on local GPUs or with NIMs hosted on build.nvidia.com, and it can also fetch documents from HTTP and HTTPS URLs. For production workloads, the project is designed to be deployed on Kubernetes with a Helm chart.

The project is also referred to as NVIDIA Ingest in some NVIDIA product materials. The default branch tracks active development, so stable releases are maintained on separate version branches.

Key features

  • 01Splits documents into pages and classifies text, tables, charts and infographics.
  • 02Runs OCR to convert extracted content into a defined JSON schema.
  • 03Computes embeddings for extracted content and stores them in LanceDB.
  • 04Can use NVIDIA NIM microservices or a range of local models.
  • 05Fetches documents from HTTP and HTTPS URLs with configurable request limits.
  • 06Provides a Helm chart for deploying the pipeline on Kubernetes.

Use cases

  • Preparing PDFs with tables and charts for a retrieval-augmented generation chatbot.
  • Building a multimodal document search index over enterprise reports.
  • Running large-scale document ingestion on a Kubernetes cluster with GPUs.
  • Extracting structured text from web-hosted documents for downstream LLM applications.

Quick start

  1. 1

    Follow the quickstart installation steps in the nemo_retriever subfolder

  2. 2

    Run the Python ingestion example from the README (create_ingestor, extract, embed, ingest)

Things to know

  • The default main branch is in active development and may be ahead of the latest supported release; use the 26.08 branch for stable use.
  • The in-development library setup is aimed at small workloads of fewer than 100 PDFs; production use requires Kubernetes and Helm deployment.
  • Hosted models run on build.nvidia.com or NIM endpoints, which need an NVIDIA API key.
  • The optional Adobe SDK for PDF extraction is governed by a separate license agreement.

Topics

#document-extraction#ocr#rag#embeddings#pdf-parsing#nvidia-nim#kubernetes#vector-database

FAQ

What is NeMo-Retriever?
A scalable pipeline that extracts text, tables, charts and images from documents into structured JSON, using NVIDIA NIM microservices and models, for RAG applications. NeMo Retriever Library is a document content and metadata extraction framework from NVIDIA. It splits documents into pages, classifies artifacts such as text, tables, charts and infographics, and runs OCR to produce a defined JSON schema. The pipeline can then compute embeddings for the extracted content and store them in LanceDB, preparing data for retrieval-augmented generation.
Is NeMo-Retriever free to use? What license does it have?
NeMo-Retriever is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is NeMo-Retriever still maintained?
Actively maintained. The last commit to NeMo-Retriever was on Oct 19, 2026.
What can I use NeMo-Retriever for?
Preparing PDFs with tables and charts for a retrieval-augmented generation chatbot. Building a multimodal document search index over enterprise reports. Running large-scale document ingestion on a Kubernetes cluster with GPUs. Extracting structured text from web-hosted documents for downstream LLM applications.
How do I install NeMo-Retriever?
Follow the quickstart installation steps in the nemo_retriever subfolder Run the Python ingestion example from the README (create_ingestor, extract, embed, ingest)
Where can I get NeMo-Retriever?
The source code and installation instructions are on GitHub: https://github.com/NVIDIA/NeMo-Retriever

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects