Knowledge Graph Extraction Agent

Experiments that use LLMs to extract entities and relationships from text or a live conversation, then visualise the resulting graph.

Included experiments

• knowledge-graph-extraction-agent.ipynb — extracts a graph from static text.
• conversational-kg-extraction-agent.ipynb — maintains a graph as a conversation develops.
• conversational_kg_extraction_agent.py — script version of the conversational flow.
• Generated graph visualisation — a static HTML artifact that opens locally in a browser.

Setup

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Create a .env file with the provider key used by your chosen implementation, for example:

DEEPSEEK_API_KEY=your_key_here

The notebooks may use a different provider configuration; inspect the selected notebook before running it. Never commit provider keys.

Run

Open either notebook in Jupyter and run its cells, or execute:

python conversational_kg_extraction_agent.py

The conversational flow exits with /exit and emits a graph visualisation.

Data safety

Use synthetic or non-sensitive conversations for demos. Text is sent to the configured LLM provider and generated HTML may contain the entities and relationships it extracted.

License

MIT