# open-extract

A schema-based extractor that pulls structured data from documents and the web into JSON or Markdown with a single tool call, for use inside AI agent workflows.

open-extract is a data-extraction service aimed at AI agents and agentic workflows built with frameworks such as LangGraph, AG2, and CrewAI. Given one or more predefined schemas, each a set of key-value pairs describing what to pull out, it identifies and extracts the matching data from one or many documents or web pages in a single API call, returning the result as JSON or Markdown.

It supports multiple schemas and documents at once without requiring a vector database or page numbers, and caches previously extracted schemas so repeat extractions do not reprocess the source material. Because it works with any model provider, open or closed source, a project is not locked into one vendor. Example use cases in the documentation include pulling metrics from financial reports, categorizing customer feedback, extracting methodology from research papers, and pulling terms from legal contracts.

Website: https://github.com/velocitybolt/open-extract