Chunkr

A parsing interface for retrieval augmentation and document pipelines, which the official says can convert PDFs, images and tables into structured data usable by models, and provides layout analysis, OCR and structured extraction.

AI Document Generator#Document Parsing#OCR Recognition#RAG Data

Updated: 2026-10-26

What is Chunkr?

Chunkr's official website defines itself as an intelligent document interface. Its job is to turn complex documents into high-quality data: parsing PDFs, images and tables, and outputting HTML, Markdown or JSON that can be directly fed into large models. It provides both a hosted cloud interface and an open-source self-deployable version on GitHub; the official clearly states that the open-source version uses community and open-source models, while the cloud interface runs on self-developed models, and the two are different in terms of accuracy and reliability.

Who is it for?

Engineering teams working on Retrieval-Augmented Generation (RAG) whose recall performance is hampered by the quality of PDF parsing.
Data teams that need to batch extract fields from contracts, financial reports and research papers and keep track of the sources.
Financial and government-enterprise technology teams that have requirements for not storing data on the internal network and must perform self-deployment.
Backend developers who need to process spreadsheet documents and require native parsing capabilities.

Key features

Layout Analysis

Identifies document structure and restores reading order. The official lists it as a core capability alongside OCR and semantic chunking.

OCR and Coordinate Boxes

OCR recognizes text and provides the position information of text blocks, making it easy to verify by referring back to the original page.

Structured HTML and Markdown

Convert PDFs, PPTs, Word documents and images into structured HTML or Markdown, officially positioned as retrieval augmentation and chunks directly usable by large models.

Structured Extraction

Extract fields according to a given structure, which the official website lists alongside OCR and citation positioning in its page descriptions.

Citation Positioning

The official states that the output includes citations to facilitate verification of the source of the extraction results.

Two paths: open-source and cloud

The open-source version on GitHub can be self-deployed using Docker Compose; the official service also offers a cloud interface and an enterprise version for high-security scenarios, supporting local or VPC deployment.

How to use Chunkr

First, test your most challenging set of documents for evaluation, then decide whether to use the open-source version or the cloud interface.
the official documentation also distinguishes between the accuracy of the two.
The official open-source version does not include native Excel parsing.
If you need to process spreadsheets, consider the cloud interface or the enterprise version.
Self-deployment requires a Docker environment, and the official repository provides three startup methods for GPU, pure CPU, and Mac ARM.
The open-source version is licensed under AGPL.
Please review the license requirements before commercial use.
the terms for the cloud API and enterprise version are subject to the official website.
Be sure to retain the citation positions in the extraction results and conduct manual sampling checks.
the accuracy may change as the models are updated.
The official website also has a Documentation Center and a Discord community.
If you encounter access issues, you can check the documentation first and then ask questions.