JupyterLite — Python notebooks, no install

An experiment: a full Jupyter notebook running the rete-graph Python client entirely in your browser tab. The kernel is Pyodide (CPython on WebAssembly) inside JupyterLite; the package installs from PyPI as a PyEmscripten wheel; remote .rete graphs are queried over HTTP range requests from inside the tab. Nothing to install, no server anywhere in the stack — not even for Python itself.

Two notebooks are bundled:

  • Tour: query graphs → — the client end to end: lazy remote graphs, SPARQL → pandas, embedded starter queries.
  • Build: anatomy of a .rete → — author a small KG with an ontology in rdflib, build a .rete from it, and dissect every part of the file: the dictionary (terms stored once, integer-id triples), the triple indexes, the pyramid (built with the types algorithm, so the summary is the class structure), the Dataset Card, and the embedded example queries — ending with OWL 2 QL reasoning inferring what was never asserted, and an exported file in the notebook's file browser.

The tour notebook, rete-graph.ipynb, walks through the whole client with explanations between the cells:

  1. %pip install rete-graph pandas — the PyEmscripten wheel resolves straight from PyPI;
  2. open the BOE legal graph lazily (447k triples, 6.9 MB on Cloudflare R2) and watch stats() count how few bytes the open actually fetched;
  3. SPARQL → pandas — query_df() renders results as real DataFrame tables, like the most-citing laws in the corpus;
  4. the dataset documents itself — the starter queries embedded in the file's Dataset Card, listed as a table and then run as-is;
  5. the schema profile as a table, and
  6. building a .rete in the browser with the lazy Builder, card and embedded example included.

Run it top to bottom with Shift+Enter; the first cell takes a moment (it boots Pyodide and downloads the wheel). Everything is editable — change the queries, point rete.open() at any .rete URL whose host sends CORS + Range (see Hosting).

Notes and limits, honestly: the kernel runs in a web worker, which is exactly where the client's synchronous XHR range reads are allowed — so remote opens work here, while SERVICE federation and threads stay disabled (see the Python API's browser notes). The Pyodide runtime loads from a CDN on first start, so the page needs network. Wheels are per-Pyodide-ABI; this deployment pins a JupyterLite release whose Pyodide matches the wheels published from this repo.

Source & issues: https://github.com/caviri/rete · © 2026 Carlos Vivar Ríos, released under the Apache License 2.0. Datasets keep their own licenses (BOE: © Agencia Estatal Boletín Oficial del Estado, reuse conditions).