#3030th of 67 in RAG, knowledge and searchFor data and documents, documents and ocr
hyper-extract
CLI that extracts knowledge graphs from documents using LLMs and YAML templates
- Stars
- 4.2k
- License
- custom license
- Last commit
- Nov 2026
- Last release
- Nov 2026
- Language
- Python
618 on 25 Apr 20264,242 on 14 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
Hyper-Extract is a Python CLI (`he`) and library that parses documents such as PDF, Word, HTML and Markdown into structured knowledge: lists, graphs, hypergraphs and temporal or spatial graphs. It ships 80+ YAML templates and several extraction methods, including GraphRAG, LightRAG and KG-Gen. Results can be searched, visualized, exported to Obsidian, or exposed over an MCP server.
Who it is for: Engineers turning documents into knowledge graphs from the command line
Strengths
- 80+ YAML templates across finance, legal, medical and general domains need no code
- Incremental feed with per-source provenance, rollback and re-feed of updated documents
- Works with OpenAI, Anthropic, Gemini, DeepSeek, Bailian or local vLLM endpoints
- Exports to Obsidian, GraphML, CSV, JSON-LD and Cypher; includes an MCP server
Weaknesses
- Requires a model with tool/function calling; extraction quality depends on that model
- Anthropic, Gemini and DeepSeek have no embeddings path, so a second embedder is needed
- PDF, Word and other non-text formats need the optional ingest extra
- Hosted APIs bill per page extracted; local vLLM use needs a GPU
What it needs
- GPU optional
- Needs LLM provider API or local vLLM, OpenAI-compatible embedding endpoint, uv or pipx
- Models: OpenAI, Anthropic, Google Gemini, DeepSeek, Alibaba Bailian
What changed
From the release notes. No flag means the note does not say so, not that the upgrade is safe.
Known problems
Reported, then fixed by a linked change
- MCP search crashes on graph_rag 3-tuples ↗ (opens in a new tab)0 reactions · closed 13 Sept 2026Fixed by fix: normalize search hits and share KA GraphML/CSV export adapters ↗ (opens in a new tab)
- Unforgiving Pydantic Validation Error ↗ (opens in a new tab)0 reactions · closed 18 Jul 2026Fixed by fix: tolerate LLM parse failures during extraction (#45) ↗ (opens in a new tab)
- <400> InternalError.Algo.InvalidParameter: Value error, batch size is invalid, it should not be larger than 10.: input.contents ↗ (opens in a new tab)0 reactions · closed 18 Jun 2026Fixed by fix: cap embedding batch size for OpenAI-compatible providers (#33) ↗ (opens in a new tab)
Titles are the reporters' own words, taken from the project's GitHub issues and ranked by reactions. We have not tried to reproduce any of them.
Also in RAG, knowledge and search
See all 67| Rank | Project | Score |
|---|---|---|
| 1 | Open NotebookSelf-hosted NotebookLM alternative for research notebooks, chat, and podcasts | 92 out of 100 |
| 2 | LightRAGGraph-based RAG framework with a REST server and WebUI | 92 out of 100 |
| 3 | RAGFlowRAG engine with document parsing, agentic retrieval and citations | 90 out of 100 |
| #4 | deeptutorTutoring workspace with knowledge bases, guided learning, memory and quizzes | 90 out of 100 |
| #5 | MaxKBEnterprise knowledge-base agent platform with RAG, workflows and MCP tools | 89 out of 100 |