Docetl

DocETL is an open source toolkit for building LLM powered pipelines that extract, transform and link knowledge from unstructured documents. It is built by the EPIC Data Lab at UC Berkeley and presents itself as a document ETL platform that turns raw text into structured data. The project is grounded in published research. The site lists papers on agentic query rewriting and evaluation for complex document processing, on steering semantic data processing with an interface called DocWrangler, and on multi objective agentic rewrites for unstructured data processing, which appeared at venues including VLDB and UIST. That research focus means the tool is oriented toward accuracy and cost tradeoffs when language models process large collections of documents, rather than a single document at a time. The site links to documentation, the GitHub repository, a showcase of examples, a browser based playground and a Discord community. It suits data engineers, researchers and analysts who want to define document processing pipelines in code or explore them interactively before running them at scale. DocETL is distributed as open source software, and the site publishes no paid plans or prices. Running pipelines depends on whichever language model provider the user connects, so model usage costs are set by that provider rather than by DocETL.