A clear, first-hand explanation of rewriting Datasette Lite to run the full server-side Python app in the browser via service workers and Pyodide, presenting a transferable architecture pattern for WebAssembly preview environments.
Background
Datasette Lite, first launched in 2022, originally ran the server-side Datasette app in a Pyodide web worker and injected the resulting HTML into the page with innerHTML. That let basic pages work, but script tags in innerHTML do not execute, so JavaScript-dependent features and extra CSS or JavaScript assets failed.
Solution
The rewrite registers a service worker for all of lite.datasette.io; an iframe requests normal Datasette URLs prefixed with /app//, and the worker routes each request through the parent page to a dedicated Pyodide web worker. The per-tab session prefix prevents tabs sharing the domain-wide service worker from colliding when each URL asks for different plugins or settings. A web worker is still required because, the author says, Pyodide inside a service worker cannot retain state beyond roughly 30 seconds depending on the browser. Carefully composed URLs can install PyPI or wheel-file plugins, load databases, JSON, CSV, SQL or metadata, and select a released Datasette version or a GitHub ref; the author uses these links to demonstrate bugs and work-in-progress features. A later ?sqlite= parameter resolves one of 163 WebAssembly-compiled pysqlite3 wheels, making it possible to test Datasette against specific SQLite releases, while the author notes non-SQLite databases remain harder and the system has many moving parts.
Takeaway
The author argues the architecture is not Datasette-specific: any Python app could use WebAssembly and Pyodide for cheap preview and debugging environments, with C and Rust extensions increasingly compilable and PEP 783 allowing Wasm wheels on PyPI.
Import AI 481 summarizes recent work on AI R&D subcontracting, AI-oriented wet labs, covert sabotage during self-improvement, electromagnetic-signal models, and edge computer vision…
A concise Latent Space episode recap that captures BFL's arguments for multimodal visual intelligence and a few practical engineering lessons, but without enough technical depth or...
Comments
Comments use a GitHub account and are moderated. Comment rules