Unstructured
Unstructured is an open-source ETL library for turning complex documents into clean, structured output for language models, with a commercial hosted API and enterprise platform. The README says the open-source library “is, and will stay, completely free”.
Repository snapshot (GitHub API, 2026-11-19)
| Item | Detail |
|---|---|
| Repo | Unstructured-IO/unstructured, Apache-2.0 |
| Created | 2022-09-26 |
| Activity | last push 2026-11-17; latest release 0.27.37 (2026-11-06) |
| Size | 15,767 stars, 1,403 forks |
Open-source library (README)
- Pre-processes images and text documents (PDF, HTML, Word and more) through partition functions such as
partition_pdf,partition_textand an auto-detectingpartition(). - Modular install: plain text, HTML, XML, JSON and email need only
pip install unstructured; other types need extras such asunstructured[docx,pptx]. PDFs and images also need poppler-utils and tesseract-ocr.
Hosted product
The README promotes a hosted Transform v2 API and an enterprise Platform (partitioning, enrichment, chunking, embedding for production workflows), and claims higher table accuracy and less invented content than the library. Those figures are the vendor’s and were not tested here.
Where it fits
Unstructured and docling solve the same ingestion step for RAG; Unstructured has a managed option, Docling is fully open source. Output typically goes on to vector-embeddings and a store such as opensearch or azure-ai-search.
Sources
- Unstructured README and GitHub API (2026-11-19)