ai-knowledge-graph

What it is

AI Powered Knowledge Graph Generator. This system processes unstructured text to extract Subject-Predicate-Object triplets using an LLM, then visualizes the relationships as an interactive knowledge graph.

How it works

• Text is chunked into overlapping segments to fit LLM context.
• Each chunk is processed to extract SPO triples via an LLM.
• Entity standardization can be enabled to normalize entity names across chunks.
• Relationship inference can be enabled to infer additional connections, including transitive and lexical-similarity based links.
• Visualization is produced with PyVis, producing an interactive HTML graph with color-coded communities and node sizing.
• Supports OpenAI-compatible endpoints via configurable LLM settings; standard components include prompts, configuration loading, and a visualization pipeline.

Getting started

• Quick Start instructions are in the repository README:
  • Clone this repository
  • Install dependencies: pip install -r requirements.txt
  • Configure your settings in config.toml
  • Run the system:

python generate-graph.py --input your_text_file.txt --output knowledge_graph.html

Or with UV:

uv run generate-graph.py --input your_text_file.txt --output knowledge_graph.html

Or installing as a module:

pip install --upgrade -e .
generate-graph --input your_text_file.txt --output knowledge_graph.html

• Configuration is done in config.toml with sections for llm, chunking, standardization, and inference.

Recent releases

• 0.6.0 v0.6.0 (2025-03-30): Moved prompts to a dedicated module file; added error handling for malformed SPO objects; removed self-referential loops.
• 0.6.1 v0.6.1 (2025-05-12): Fixed encoding issue on Windows.
• 0.6.2 v0.6.2 (2025-12-28): Added filter for inferred vs non-inferred predicates; added utility to convert from (truncated in README).
• 0.6.3 v0.6.3 (2025-12-28): Introduce PromptFactory and modular prompts package.

Traction

• Stars: 2781
• Forks: 382
• Open issues: 13

Behind the repo

• The project appears under the user/organization robert-mcdermott and uses a license Apache-2.0. It targets Python 3.11+ and lists dependencies in a requirements file. The repository includes a src/knowledge_graph package with modules for configuration, llm interaction, entity standardization, prompts, and visualization, along with a graph template.

Caveats

• License: Apache-2.0
• Created: 2025-03-23
• Last push: 2025-12-28
• Requires Python 3.11+ and dependencies from requirements.txt or uv tooling as noted in the README.