Turn any document into a structured knowledge graph

Knwler is a lightweight, single-file Python tool that extracts structured knowledge graphs from documents using AI. Feed it a PDF or text file and receive a richly connected network of entities, relationships, and topics — complete with an interactive HTML report and exports ready for your favorite graph analytics platform.

Built for compliance teams, legal departments, research analysts, and anyone who needs to rapidly understand the structure hidden inside dense documents.

Choose between OpenAI for maximum speed, or Ollama for fully offline, air-gapped operation. Qwen 2.5 at 3B–14B parameters delivers strong results locally. You can even switch backends between runs and incrementally augment the same graph.

The pipeline analyzes a sample of your document and infers the optimal entity types and relation types — no manual ontology engineering required. You can also supply a schema if you wish. A schema is a set of types of entities (person, concept, location...) and relations (knows, has_accepted, has_signed...).

Language is auto-detected on every run. All prompts (summarization, extraction, community labeling) and all console/UI output are localized. Adding a new language is as simple as extending a single JSON file.

Re-run on new documents or updated schemas and the existing graph is augmented rather than rebuilt. Entity descriptions from multiple sources are intelligently consolidated via LLM-powered summarization.

The Louvain algorithm automatically discovers clusters of related entities and an LLM labels each community with human-readable topics — giving you instant thematic insight into the document's structure.

Export a single HTML file with interactive Cytoscape.js network visualization, entity index, topic overview, and rephrased text chunks — shareable without any server or dependencies.

• JSON — the canonical output; import into Neo4j, Memgraph, SurrealDB, or generate vector embeddings
• GML / GraphML — open directly in yEd, Gephi, or any standards-compliant graph tool
• HTML — standalone interactive report

Every LLM call is hashed and cached locally. Re-generating reports, tweaking export settings, or re-running with a different schema costs zero additional API calls.

Each text chunk is rephrased for readability alongside the original, making the report accessible to non-expert stakeholders while preserving full traceability to source text.

Handles PDF-to-text extraction (via PyMuPDF) as well as plain text and Markdown files. Extracted text is cached to avoid redundant PDF parsing on subsequent runs.

A single Python file (~2,000 lines), managed via uv with minimal dependencies. No database, no backend server, no Docker required.

# Install dependencies
uv sync

# Run with OpenAI
uv run main.py --openai -f document.pdf

# Run fully local with Ollama
uv run main.py -f document.pdf

# Re-export HTML only (no LLM calls)
uv run main.py --html-only

Tip: When running Ollama locally, launch it via CLI with parallel processing for best throughput:

OLLAMA_NUM_PARALLEL=8 ollama serve

Adjust the number based on your machine specs (8 is suitable for a Mac M4 Pro with 64 GB RAM).

# EU AI Act (English)
uv run main.py --openai \
  --url "https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=OJ:L_202401689" \
  -f samples/EUAI.pdf

# NIST AI Risk Management Framework
uv run main.py --openai \
  --url "https://nvlpubs.nist.gov/nistpubs/ai/nist.ai.100-1.pdf" \
  -f samples/Nist.pdf

# Belgian Civil Code (Dutch — auto-detected)
uv run main.py --openai \
  --url "https://www.ejustice.just.fgov.be/cgi/article_body.pl?language=nl&pub_date=2022-07-01&caller=list&numac=2022032058" \
  -f samples/BurgerlijkBoek5.pdf

# Deloitte Sustainability Report (German — auto-detected)
uv run main.py --openai \
  --url "https://www.deloitte.com/de/de/legal/publikationen.html" \
  -f examples/Deloitte/Deloitte-Nachhaltigkeitsbericht-2024.pdf

The raw JSON output is designed for downstream integration:

• Import into Neo4j / Memgraph / SurrealDB — entities and relations map directly to nodes and edges
• Generate vector embeddings — use entity descriptions for semantic search
• Feed into n8n workflows — connect document intelligence to CRM, alerting, or reporting pipelines without code
• Visualize in yEd or Gephi — open the GML/GraphML export for advanced layout and analysis

You can find example reports and raw graph data in diverse languages in the examples directory.

Everything language related sits in the languages.json and this contains both the language-specific prompts as well as the text used for console output. You can easily add additional languages, simply ask Copilot, Gemini or any AI to translate the JSON.

If you run the process in your terminal the code will look for the usual OPENAI_API_KEY. You can assign it explicitly via a terminal export

export OPENAI_API_KEY=...

or in the code (look for os.environ.get("OPENAI_API_KEY", "")).

Ollama is just a convenient local LLM service, you can use LMStudio or any other service. The default model is Qwen 2.5 but here as well, experiment and see what works best for you. We have done lots of benchmarks and bigger models are not better, sometimes quite the opposite. Small models of 3 or 7 billion parameters will be fine and a lot faster. Thinking, in particular, is really standing in the way of graph extraction. Whatever you do, don't enable thinking and don't use advanced MOE models.

The information extracted by Knwler is generated via machine learning and natural language processing, which may result in errors, omissions, or misinterpretations of the original source material. This tool is provided "as is" for informational purposes only. Users are advised to independently verify any critical data against original source documents before making business, legal, or financial decisions.

Built by Orbifold Consulting and inspired by Knwl.