LangExtract

Google-released open-source document extraction tool positioned as a free alternative to enterprise extraction solutions. Handles structured and unstructured documents.

What It Is

LangExtract is Google's open-source entry into the document extraction space. It parses documents (PDFs, forms, invoices, contracts, etc.) and extracts structured data - competing with enterprise solutions that typically cost $100K+ annually (AWS Textract, Azure Document Intelligence, Kofax, ABBYY).

Current primary source framing is narrower and more useful than the viral tweet: LangExtract is a Python library for extracting structured information from unstructured text with precise source grounding and interactive visualization. The package is Apache-2.0 and supports Gemini, OpenAI, Ollama/local models, and custom providers through examples and provider docs. Source: GitHub google/langextract, 2026-06-30

Current Version Snapshot

Source: PyPI JSON; GitHub API; git ls-remote, 2026-06-30

Relevance to Kevin's Stack

Fits into the same pipeline layer as MarkItDown - Universal File-to-Markdown Converter (Microsoft's file-to-Markdown converter) and Defuddle - Web Content Extraction (web page content extraction). Where MarkItDown converts files to Markdown for LLM consumption and defuddle extracts from web pages, LangExtract focuses on structured data extraction from complex documents - invoices, forms, tables with irregular layouts.

For the wiki's ingestion pipeline: LangExtract handles documents where MarkItDown's generic conversion might lose structural meaning (e.g., extracting line items from invoices, fields from government forms).

Routing order in Kevin's stack: use MarkItDown - Universal File-to-Markdown Converter/Defuddle - Web Content Extraction for format-to-text or web cleanup, MinerU for hard multimodal/layout-heavy PDFs, LiteParse when the Agent Skill + parser route is the best fit, then LangExtract when the output needs grounded fields, schema consistency from examples, and a verification UI rather than only Markdown.

Capabilities

From the launch thread, LangExtract extracts structured data from messy text and grounds every extracted field to its exact location in the source document, so each value is traceable back to the original. It handles 100+ page documents, generates an interactive HTML view for verifying extractions, and runs against both Gemini and local models. It targets work previously handled by regex and fragile parsing, custom NER pipelines, expensive extraction APIs, and manual data entry. Named business use cases: invoice and PO processing, customer feedback to themes to actions, support tickets to root causes to SOP updates, legal documents to clauses to risk summaries, and financial reports to KPIs to exec briefs. Source: X/@mdancho84, 2026-04-02

The reviewed local image shows the README table of contents and confirms the exact capability claims worth preserving: source grounding, reliable structured outputs from user-defined examples, long-document extraction, custom model providers, OpenAI support, and local LLM/Ollama support. It also showed an older v1.1.1 badge; the current package authority is PyPI/GitHub 1.5.0, so screenshots should not be used as version truth. Source: local asset /wiki/assets/x-bookmarks/2039726865775051135/image-01.jpg, 2026-06-30

Signal

• ~6,971 combined engagement - high interest in open-source alternatives to expensive enterprise extraction tools

---

Timeline

• 2026-04 | LangExtract shared on X by @mdancho84. "Google just released LangExtract: Open-source. Free. Better than $100K enterprise solutions." Source: X/@mdancho84
• 2026-06-30 | Deep-reviewed the launch thread and local README screenshot, then refreshed source authority to PyPI langextract==1.5.0, GitHub tag v1.5.0, and repo HEAD fef3e7db723e87d9cdd11dfeda219bf4fa269350. Decision: update the tool page and document-extraction routing; no new local skill until repeated LangExtract project work needs executable recipes. Source: X/@mdancho84, 2026-04-02; Source: GitHub google/langextract, 2026-06-30
• 2026-05-17 | Added to wiki tool registry. Source: X bookmark absorption, 2026-05-17
• 2026-06-01 | Capabilities expanded from enriched launch thread 2039726865775051135: source grounding to exact location, interactive HTML verification, 100+ page documents, Gemini plus local models, and named business use cases. Source: compiled from enrichment, 2026-06-01