Hyper-Extract: An Open-Source CLI for Turning Documents into Structured Knowledge Graphs

Your documents are full of facts. Most tools still treat them like blobs of text.
Introducing Hyper-Extract — an open-source, LLM-powered CLI that turns unstructured documents into structured, queryable Knowledge Abstracts.
Parse → Extract → Query → Visualize. All from the command line.
What makes it different:
🔹 One-command workflow — go from raw document to structured, visualizable output without switching tools
🔹 8 knowledge structures — from simple lists and Pydantic models to graphs, hypergraphs, and spatio-temporal graphs
🔹 80+ YAML templates — ready-made presets for finance, legal, medical, industry, and general use cases
🔹 Pluggable extraction engines — GraphRAG, LightRAG, Hyper-RAG, KG-Gen, and more
🔹 Run it fully local — vLLM + Qwen3.5-9B + bge-m3 for extraction and embeddings, no external API required
Fully open-source under Apache 2.0 — built for builders who want structure, not black boxes.
🔗 Link in the comments 👇

This is a factual product-announcement post for a real open-source project. Most concrete claims (repo existence, features, licensing) are independently verifiable and check out against the project's GitHub, PyPI, and docs. The language is promotional in tone (as expected for a launch post) but not misleading about the tool's actual capabilities.

Confirmed Accurate

• Hyper-Extract is a real, open-source LLM-powered CLI and Python framework for turning unstructured text into structured "Knowledge Abstracts."
• It follows a Parse → Extract → Query → Visualize-style workflow via CLI commands (e.g., he parse, he search, he show).
• It supports 8 knowledge structure types ("Auto-Types"), ranging from simple Lists/Models to Graphs, Hypergraphs, and Spatio-Temporal Graphs.
• It ships 80+ YAML templates covering domains including Finance, Legal, Medical, TCM (Traditional Chinese Medicine), Industry, and General use.
• It supports multiple pluggable extraction engines/methods, including GraphRAG, LightRAG, Hyper-RAG, and KG-Gen, with 10+ total listed.
• The project is released under the Apache 2.0 license, and is available via PyPI (pip install hyperextract).
• It supports incremental extraction — feeding new documents to expand an existing knowledge base without reprocessing everything.
• It includes an MCP server component (he-mcp) for exposing knowledge abstracts to MCP-capable assistants like Claude Desktop.
• Qwen3.5-9B is a real, currently available open-weight model from Alibaba's Qwen team (released around late February/early March 2026), and bge-m3 is a real, widely used open embedding model — so local-only deployment via vLLM + these models is a plausible, verifiable technical setup.

Not Fully Accurate / Promotional

• The framing "Your documents are full of facts. Most tools still treat them like blobs of text." is marketing copy/opinion, not a factual claim, and sets up the tool as more novel than the competitive landscape suggests — several other tools (GraphRAG, LightRAG, KG-Gen, and Google's own structured-extraction libraries) already address similar structured-extraction needs, and Hyper-Extract itself builds on and wraps these existing engines rather than inventing the underlying extraction paradigms from scratch.
• "No external API required" is accurate only for the specific local stack described (vLLM + Qwen3.5-9B + bge-m3); the tool as a whole also explicitly supports OpenAI, Anthropic, and other API-based providers per its own documentation, so "fully local" is one optional mode, not the tool's only or default mode.
• The "10+ pluggable extraction engines" claim in the post is understated compared to some of the project's own materials (which also list iText2KG, HypergraphRAG, Cog-RAG, etc.) — not misleading, just a slightly conservative number rather than an inflated one.

Not Independently Verifiable

• Real-world extraction quality, accuracy, or performance claims are not made explicit in this post, so there's nothing to verify on that front — it's mostly a feature list rather than a performance claim.
• The project is described elsewhere as pre-1.0 (e.g., v0.3.x), meaning interfaces and extraction quality may still be actively changing; this maturity context isn't mentioned in the post itself and is worth keeping in mind if evaluating it for production use.