unstructured-io/unstructured is an open source library that breaks unstructured data such as PDF, Word documents, HTML, and images into meaningful elements and converts them into structured text. The latest release on GitHub is 0.27.32, and what changed in this release is not recorded here.
What this library does is split documents into element-level units instead of reading them whole, organizing them as data to feed into language models. The README explains that it collects and preprocesses unstructured data such as images, PDF, HTML, and Word documents to simplify LLM data processing workflows. It provides modular functions and connectors that help efficiently convert unstructured data into structured results across various platforms. The repository describes this purpose as an ETL (extract, transform, load) library. The license is Apache-2.0, GitHub stars number 15583, and the repository was created on September 26, 2022. The split output is used for RAG ingestion.
However, this record contains neither the changelog for 0.27.32 nor any figures comparing processing performance. The only number left in the record is a momentum score of 3.3 (out of 10), which reads low, and the technology, quality, and novelty scores are empty. The repository description also points to an enterprise platform product covering partitioning and enrichment, chunking, and embedding, but how much of that the library itself provides cannot be confirmed from this record. With such thin evidence, it is more appropriate for now to flag it as a candidate for document preprocessing in front of RAG.