Repolode
Submit
lumina-ai-inc

lumina-ai-inc/chunkr

Self-hosted document intelligence API that turns PDFs, Office files and images into LLM-ready chunks with layout analysis and OCR.

View on GitHubWebsite
Stars
4,157
Forks
278
Open issues
19
Last commit
2 months ago

What is it?

Chunkr is a document processing service that takes PDFs, PowerPoint files, Word documents and images and breaks them into structured chunks for retrieval-augmented generation and other LLM pipelines. It combines layout analysis, OCR with bounding boxes, and vision-language model processing to output structured HTML and Markdown.

The open-source AGPL release runs locally through Docker Compose, with GPU and CPU configurations and a web UI on port 5173 alongside an API on port 8000. It uses community and open-source models, and LLM access is configured through a models.yaml file or environment variables that accept any OpenAI-compatible endpoint, such as OpenAI, OpenRouter, Google AI Studio, or self-hosted vLLM and Ollama servers.

The project is dual-licensed under AGPL-3.0 and a commercial license. Its maintainers also offer a hosted Cloud API that runs proprietary models, which they describe as higher accuracy than the open-source version.

Key features

  • 01Layout analysis that identifies document structure before chunking.
  • 02OCR with bounding boxes for text location on each page.
  • 03Output as structured HTML and Markdown.
  • 04Supports PDF, PPT, Word documents and images.
  • 05Works with any OpenAI-compatible LLM endpoint, including self-hosted vLLM and Ollama.
  • 06Runs locally with Docker Compose in GPU or CPU mode.

Use cases

  • Preparing PDF and Office documents as chunks for a RAG pipeline.
  • Extracting text and layout from scanned files with bounding-box coordinates.
  • Running document parsing locally where data must stay on your own infrastructure.

Quick start

  1. 1

    Install Docker and Docker Compose (NVIDIA Container Toolkit optional for GPU)

  2. 2

    Clone the repository

    git clone https://github.com/lumina-ai-inc/chunkr && cd chunkr
  3. 3

    Copy the example env and models config files

    cp .env.example .env && cp models.example.yaml models.yaml
  4. 4

    Start the services (GPU deployment)

    docker compose up -d

Things to know

  • Open-source version uses community models and is less accurate than the proprietary Cloud API.
  • Requires an LLM endpoint or API key for LLM features.
  • Licensed under AGPL-3.0, which has copyleft obligations unless you buy a commercial license.
  • Excel support is not included in the open-source release.

In collections

PDF, OCR & document tools

Topics

#document-parsing#ocr#rag#llm#layout-analysis#pdf#docker#vision-language-model

FAQ

What is chunkr?
Self-hosted document intelligence API that turns PDFs, Office files and images into LLM-ready chunks with layout analysis and OCR. Chunkr is a document processing service that takes PDFs, PowerPoint files, Word documents and images and breaks them into structured chunks for retrieval-augmented generation and other LLM pipelines. It combines layout analysis, OCR with bounding boxes, and vision-language model processing to output structured HTML and Markdown.
Is chunkr free to use? What license does it have?
chunkr is open source under the AGPL-3.0 license. Check the license terms before using it in commercial products.
Is chunkr still maintained?
Updated within 6 months. The last commit to chunkr was on Sep 9, 2026.
What can I use chunkr for?
Preparing PDF and Office documents as chunks for a RAG pipeline. Extracting text and layout from scanned files with bounding-box coordinates. Running document parsing locally where data must stay on your own infrastructure.
How do I install chunkr?
Install Docker and Docker Compose (NVIDIA Container Toolkit optional for GPU) Clone the repository: git clone https://github.com/lumina-ai-inc/chunkr && cd chunkr Copy the example env and models config files: cp .env.example .env && cp models.example.yaml models.yaml Start the services (GPU deployment): docker compose up -d
Where can I get chunkr?
The source code and installation instructions are on GitHub: https://github.com/lumina-ai-inc/chunkr

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

Related projects