Wikidata-100MB

one graph, queried lazily over HTTP range - pick a backend on the left ← playground tier: 100MB · 1GB parallel reads workers data on Hugging Face ↗

The same Wikidata slice in four lossless encodings on Hugging Face. Every query below fetches only the bytes it touches - nothing is downloaded whole. Pick a backend; the example buttons load the analogous query for it. Type tables are named by Q-id (Q5=human, Q16521=taxon). A browser sends HTTP Range requests to a remote .rete file shown as a stack of tiles; only a few tiles are highlighted as fetched while most are never transferred. Caption: a 100 MB / 1 GB graph stays interactive because only a few MB cross the wire per query.

Graph

rete · SPARQL · SHACL

Lazy tasks over wikidata.rete (105 MB).

task: read:

ready (lazy)

Tables

DuckDB-WASM · Parquet

SQL over the per-type columnar entity-tables — the .parquet files (115 MB), read lazily over HTTP range.

initializing...

DuckDB

DuckDB-WASM · .duckdb

The same tables in one wikidata.duckdb file (214 MB).

read:

initializing...

SQLite experimental

sql.js-httpvfs · sql.js

SQL over wikidata.sqlite (475 MB); list/map columns are JSON text. download

read:

initializing...

Graph - SPARQL

Lazy SPARQL over the .rete graph.

Remote data isn't reachable from this browser for the current source (HF Bucket) — a range fetch returned an error/405, so the lazy Graph, Tables and SQLite backends can't fetch it here. (The rete CLI works fine: it isn't subject to CORS.) Try another host with the Source button (top-right) — HF Datasets or same-origin GitHub Pages — or cache a copy locally. You can still browse the example queries below.

Data source

Graph - SPARQL