Local Knowledge Graph

Ask a local model a question, watch it reason step by step — and see the reasoning as a knowledge graph that is used, not just drawn. Blue edges are embedding similarity — association. Green is what an exact evaluator settled — sums and unit conversions, in fractions, via mpeqs. Everything runs on your machine.

The graph is load-bearing

Every claim below is a measurement pinned in docs/claims/ and re-checked by CI on every push — the numbers cannot drift from the code.

Not a general knowledge-graph-builder for documents or RAG: the graph here is built from the model's own reasoning and fed back into it.

Run it

pip install mpe-lkg
mpe-lkg

Open http://localhost:5100. It talks to a local model through Ollama and will tell you what is missing and the one command that fixes it. The model matters more than anything else here — measured, qwen3:4b-instruct-2507 answers 82.5% of the arithmetic battery against llama3.2:3b's 40%:

ollama pull qwen3:4b-instruct-2507-q4_K_M

Ask harder questions with mode=explore (each sub-question answered by its own run) or mode=settle (explored twice, finished only when two independent runs agree), and take any run as RDF:

curl -sX POST localhost:5100/jobs -H 'content-type: application/json' \
     -d '{"query":"How many seconds are there in 23 weeks?","mode":"settle"}'
curl -s localhost:5100/jobs/<id>/rdf

More

Licence

The mpedb License 1.0 — the same licence as mpedb and MPEqs, byte for byte.

Free 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.

_{Published to PyPI as mpe-lkg — Morten Punnerud-Engelstad Local Knowledge Graph.}