From Wikidata, pre-structured
86.2%Imported by identifier, not by name, so it carries Wikidata's strengths and its gaps.
A knowledge graph made to be explored
1,063,614 entities.7,173,830 facts.One queryable graph.
The graph draws on Wikidata, Wikipedia and other open sources, with music data from open music databases, and it is built to take in your own documents. Most of it comes from Wikidata, mapped onto a small, stable set of types with identifiers that don't change. We add facts read out of Wikipedia text by a language model; each one goes through automatic type and date checks, and samples are checked by hand against the passage they came from. Every fact records its source, and we publish how each source has been checked.
A reviewed sample of the graph is available to evaluators on request, under a license agreement.
Facts come in levels of detail. Polonium, created by, Marie Curie is broad; born in Warsaw is specific. A subject layer, like music, adds its own. Her story →
From WikidataRead from text, checked by hand
What the pictures above show
A fact
Facts come in levels of detail. Polonium, created by, Marie Curie is broad; born in Warsaw is specific. A subject layer, like music, adds its own.
- polonium (chemical element), created by, Marie Curie (physicist and chemist): from Wikidata.
- radium (chemical element), created by, Marie Curie (physicist and chemist): from Wikidata.
- 7000 Curie (asteroid), named after, Marie Curie (physicist and chemist): from Wikidata.
- Marie Curie (physicist and chemist), born in, Warsaw (capital of Poland): from Wikidata.
Sources
BaseKG’s Wikidata facts name Irène’s parents and the marriage. That Marie worked at the Sorbonne was read from Wikipedia text.
- Pierre Curie (physicist), parent of, Irène Joliot-Curie (scientist): from Wikidata.
- Marie Curie (physicist and chemist), parent of, Irène Joliot-Curie (scientist): from Wikidata.
- Marie Curie (physicist and chemist), spouse of, Pierre Curie (physicist): from Wikidata.
- Marie Curie (physicist and chemist), affiliated with, Sorbonne (historic university): read from Wikipedia text, checked by hand.
Names
Many things are called Babylon. Each is its own entity, told apart by what it connects to.
- Esagila (temple of Marduk), part of, Babylon (ancient city): from Wikidata.
- North Babylon (place in New York), part of, Babylon (town in New York): from Wikidata.
- Babylon (film), has genre, period drama film (genre): from Wikidata.
- Babylon (Czech village), part of, Domažlice District (district, Czech Republic): from Wikidata.
- Babylon (translation program), is a, software (concept): from Wikidata.
Paths
Who died at the Alamo? Two lines lead straight to it. Travis is one step further out, through the battle.
- Alamo Mission (historic place), part of, San Antonio (city in Texas): from Wikidata.
- Davy Crockett (frontiersman), died in, Alamo Mission (historic place): from Wikidata.
- James Bowie (pioneer), died in, Alamo Mission (historic place): from Wikidata.
- Battle of the Alamo (battle), located in, Alamo Mission (historic place): from Wikidata.
- William B. Travis (Alamo commander), died in, Battle of the Alamo (battle): from Wikidata.
Time
Entities carry dates, so the same three people can be laid out in time.
Stories from the graph
All stories →
The International Space Station
A space station in low Earth orbit, run by five space agencies: NASA, Roscosmos, ESA, JAXA and CSA.
8 chapters
The Alamo
A mission church in San Antonio, and thirteen days in 1836 that Texas has retold ever since.
11 chapters
Marie Curie
A physicist and chemist from Warsaw who discovered two elements and won the Nobel Prize twice.
12 chapters
Gershwin and “Summertime”
A composer and pianist from Brooklyn who wrote for Broadway, the concert hall and the opera house.
10 chapters
BaseKG in one minute
Two kinds of fact
Read from text
13.8%Relations a language model reads out of Wikipedia or other text. It finds more, and it can be wrong in ways an import can't.
Each kind is checked in its own way, and the two results are never blended into one number.
What it brings
The same answer every time.
It is a source you can query directly, rather than asking a generative model and hoping. Same query, same release, same answer.
The graph is a single SQLite file. It runs on a laptop or a Raspberry Pi, and can be served to a model through a local MCP server. A subject layer, like music, is a file of its own.
Explore a sample →More than one source.
Wikidata, Wikipedia, other sources and your own documents, each fact labeled with where it came from. More sources are on the way.
Organized detail.
Every fact has a level of detail. At the first, relations are few and broad: part of, is a, created by. The second makes them specific: born in, educated at, named after. A layer for one subject goes further: music's says performed by, composed by, produced by. So polonium, created by, Marie Curie is a first-level fact, because the graph has no finer relation for a discovery.
Music is the first subject layer, joined to the graph by entity id. Your own documents can make a layer the same way.
The music layer →Checked where it matters.
If a decision rests on the data, breadth is not enough. One subject can be checked on its own: a sample of it checked by hand, with its own error table. You get a number for the part you use, not an average over a graph you don't.
How good is the data? →