Chunkr

Chunkr is an open-source document intelligence API and managed cloud service for turning complex files into data that retrieval and LLM systems can actually use. The GitHub README describes layout analysis, OCR with bounding boxes, structured HTML and Markdown output, semantic chunking, and vision-language model processing for PDFs, PowerPoint files, Word documents, images, and, on the cloud product, Excel files. For teams building RAG applications, document search, underwriting automation, legal review, or knowledge-base ingestion, Chunkr sits in the ingestion layer before embeddings and generation. The product is split between an AGPL open-source release and the managed Chunkr Cloud API. The repository says the open-source version is aimed at development, testing, transparency, and self-hosting. It uses community and open-source models. The cloud API, available at chunkr.ai, uses proprietary in-house models for higher accuracy, speed, and production reliability, with the enterprise option adding custom tuning, managed or on-prem deployment, and dedicated migration support. That distinction is important: the GitHub repo is useful for local control, while the hosted service is the more direct path for production workloads. Chunkr's feature set is centered on preserving document structure. It is designed to identify layouts, run OCR, attach bounding boxes, emit HTML or Markdown, and produce chunks that are ready for RAG and LLM pipelines. That matters for documents with tables, slides, forms, scanned pages, diagrams, and mixed visual/text regions where plain text extraction loses context. The README positions Chunkr as production-ready infrastructure rather than a one-off parser. Developers should use Chunkr when they need repeatable document processing with clear output formats and deployment flexibility. AI product teams can start with the open-source stack, compare outputs, and move to the cloud API if they need better accuracy or managed scaling. Pricing for the managed API is not fixed in the README, so buyers should check the official website for current plan details or contact the company for production and enterprise usage. The strongest fit is any RAG workflow where extraction quality controls downstream answer quality. The main thing to verify before adopting Chunkr is the quality of output on your own document set. Layout-heavy PDFs, tables, scanned pages, and slide decks can expose weaknesses in simple parsers. Chunkr is positioned for those harder cases, but every RAG team should test representative files and compare the resulting chunks against expected retrieval behavior. If the open-source version is accurate enough, it gives control and transparency. If production speed, accuracy, or Excel support matters more, the managed API or enterprise path is the better fit.