About this project

Free knowledge, made listenable.

Curio Garden is an accessibility-first experiment in turning the depth and serendipity of Wikipedia into a calm listening library. Search an article, choose a section, save your place, or carry a queue into your podcast app.

The idea

Curiosity should not depend on a screen.

Curio Garden began with a specific access need. Its founder, Seth Wilson, is visually impaired and wanted a calmer way to explore long Wikipedia articles by listening without giving up structure, source context, or control. That experience made audio the starting point—not an afterthought bolted onto a wall of text.

That origin is context, not a claim to represent everyone. Blind and low-vision people use different tools, have different preferences, and encounter different barriers; people affected by fatigue, mobility, cognitive access, or a temporary situation add still more perspectives. One person's workflow is one data point, so Curio Garden has to keep learning from people whose experiences differ from Seth's.

It also tries to honor the thing that makes Wikipedia possible: transparent sourcing, contributor history, open licenses, and the invitation to keep learning beyond this interface.

Help the garden learn

Your experience can shape what grows next.

If you use a screen reader, magnification, voice control, keyboard navigation, or another access approach, Curio Garden would like to learn what works and what gets in your way. Product feedback from every visitor is welcome, too. You do not need to share a diagnosis, and you can choose whether to volunteer for a possible short research conversation. Volunteering is not a mailing list or a commitment.

Share feedback or volunteer for research

Under the garden bed

A product, not a prototype shell.

  1. Revision-aware Wikipedia ingestion

    Articles retain their source revision, contributor-history link, citations, related topics, and media provenance while being reshaped for listening.

  2. Resilient audio delivery

    Section-level synthesis, provider-aware caching and fallback, resumable playback, and downloadable articles keep listening useful when one service is under pressure.

  3. Web and podcast, one pipeline

    The same structured article audio powers browser playback, featured-article episodes, clearly labeled AI-generated trending briefings, and private personal playlist feeds.

  4. Accessibility in the architecture

    Semantic controls, keyboard navigation, visible focus, live status updates, reduced motion, local persistence, and screen-reader-friendly structure are treated as product behavior.

Plain-language provenance

Wikipedia text and generated context stay distinct.

Article text remains tied to an exact Wikipedia revision. Curio Garden's Trending summaries and podcast scripts are generated with OpenAI from Wikimedia pageview data and linked reporting; they are labeled as AI-generated on the page, in podcast metadata, and at the beginning of the audio.

Rich context notes start from deterministic source extraction. When AI helps make a description clearer, that assistance and the model are disclosed beside the source revision. Generated material may contain errors, and readers can report a problem from each note.

Built to be explored and inspected.

Local mode works without accounts or a database for quick inspection. The full deployment adds Convex caching and storage, Clerk-backed accounts, synced libraries, private podcast feeds, analytics, and scheduled publishing. The source, tests, setup notes, and tradeoffs are available in the public repository.

View the source on GitHub (opens in new tab)