{"slug": "show-hn-visual-knowledge-graphs-using-ollama-and-embeddings", "title": "Show HN: Visual knowledge graphs using Ollama and Embeddings", "summary": "Morten Punnerud-Engelstad released mpe-lkg, a local knowledge graph tool that visualizes a language model's step-by-step reasoning using Ollama and embeddings, with blue edges for embedding similarity and green edges for exact evaluations via MPEqs. The tool, published on PyPI, runs entirely on the user's machine and supports modes like explore and settle, with RDF export. Benchmarking shows qwen3:4b-instruct-2507 answers 82.5% of arithmetic questions versus llama3.2:3b's 40%.", "body_md": "Ask a local model a question, watch it reason step by step, and see the steps drawn as a\ngraph. **Blue** edges are embedding similarity — association. **Green** is what an exact\nevaluator settled — sums and unit conversions, in fractions, via\n[mpeqs](https://github.com/punnerud/MPEqs). Everything runs on your machine.\n\n```\npip install mpe-lkg\nmpe-lkg\n```\n\nOpen [http://localhost:5100](http://localhost:5100). It talks to a local model through\n[Ollama](https://ollama.com) and will tell you what is missing and the one command that\nfixes it. The model matters more than anything else here — measured, `qwen3:4b-instruct-2507`\n\nanswers 82.5% of the arithmetic battery against `llama3.2:3b`\n\n's 40%:\n\n```\nollama pull qwen3:4b-instruct-2507-q4_K_M\n```\n\nAsk harder questions with `mode=explore`\n\n(each sub-question answered by its own run) or\n`mode=settle`\n\n(explored twice, finished only when two independent runs agree), and take any\nrun as RDF:\n\n```\ncurl -sX POST localhost:5100/jobs -H 'content-type: application/json' \\\n     -d '{\"query\":\"How many seconds are there in 23 weeks?\",\"mode\":\"settle\"}'\ncurl -s localhost:5100/jobs/<id>/rdf\n```\n\n|\n\n[Models and configuration](/punnerud/Local_Knowledge_Graph/blob/main/docs/models.md)[Embeddings from inside a model](/punnerud/Local_Knowledge_Graph/blob/main/docs/internal-layers.md)[How it works](/punnerud/Local_Knowledge_Graph/blob/main/docs/design.md)[Development](/punnerud/Local_Knowledge_Graph/blob/main/docs/development.md)[The mpedb License 1.0](/punnerud/Local_Knowledge_Graph/blob/main/LICENSE) — the same licence as\n[mpedb](https://github.com/punnerud/mpedb) and [MPEqs](https://github.com/punnerud/MPEqs),\nbyte for byte.\n\nFree of charge for every person and every organization, with one exception: a group whose revenue or valuation exceeds five billion dollars owes a one-time fee of seven US cents per device. Not an OSI-approved licence.\n\nPublished to PyPI as mpe-lkg — Morten Punnerud-Engelstad Local\nKnowledge Graph.", "url": "https://wpnews.pro/news/show-hn-visual-knowledge-graphs-using-ollama-and-embeddings", "canonical_source": "https://github.com/punnerud/Local_Knowledge_Graph", "published_at": "2026-08-12 10:15:21+00:00", "updated_at": "2026-08-12 10:42:27.499730+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-tools", "ai-products", "natural-language-processing", "developer-tools"], "entities": ["Morten Punnerud-Engelstad", "mpe-lkg", "Ollama", "qwen3:4b-instruct-2507", "llama3.2:3b", "MPEqs", "PyPI"], "alternates": {"html": "https://wpnews.pro/news/show-hn-visual-knowledge-graphs-using-ollama-and-embeddings", "markdown": "https://wpnews.pro/news/show-hn-visual-knowledge-graphs-using-ollama-and-embeddings.md", "text": "https://wpnews.pro/news/show-hn-visual-knowledge-graphs-using-ollama-and-embeddings.txt", "jsonld": "https://wpnews.pro/news/show-hn-visual-knowledge-graphs-using-ollama-and-embeddings.jsonld"}}