Nvidia Ocr

High-precision OCR for images, tables, and handwriting using NVIDIA NeMo Retriever.

Install
cmdop skills install agensi-nvidia-ocr

This skill provides Optical Character Recognition (OCR) by leveraging the NVIDIA NeMo Retriever API. It allows an AI agent to “see” and extract text from images and documents with professional-grade accuracy. The capability handles complex structures such as tables, charts, receipts, and even handwriting, returning structured text along with confidence scores and bounding box data. The main tool provided is the NVIDIA NeMo Retriever, a foundational OCR model that specializes in precision. For developers, the skill includes Python integration which facilitates Built-in handling for Base64 encoding and batch file processing, meaning agents can process multiple files efficiently. Additionally, it offers exporting functionality, allowing results to be saved locally in .txt or .md formats for easy developer access. This combination of tools ensures that an agent can not only extract but also manage and store the recognized text effectively. This is designed for tasks requiring precise data extraction from visual inputs, especially when dealing with dense or challenging text formats.

Use cases

  • Extract text from images in documents
  • Automate data entry from scanned receipts
  • Process information from tables and charts
  • Digitize handwritten notes
  • Convert image-based reports into searchable text

When to use it

  • When high-accuracy text extraction from images is required
  • For processing complex document structures like tables
  • When batch processing of multiple image files is needed
  • When confidence scores for extracted text are necessary
  • When standard LLM vision capabilities are insufficient for OCR accuracy

When not to use it

  • For tasks that do not involve optical character recognition
  • When only basic text recognition is needed without structural detail
  • If the NVIDIA NeMo Retriever API is not accessible
  • For analyzing content that is not primarily text-based
  • When offline processing without external API access is a strict requirement