Online Sources

The Sources view (🌐 in the icon rail) searches two free online libraries — Aozora Bunko and Japanese Wikisource — and imports works into your library in one click.

The view

Two tabs sit next to the heading: 青空文庫 (Aozora Bunko) and Wikisource. A single search box serves both; its behavior differs per tab, as described below. While an import is running, a spinner with an "importing N…" counter appears in the header, and the Library nav icon shows the same count.

Imported works go through the normal import pipeline and land in the Library exactly like a file import: they are tokenized, difficulty-graded, and openable in the Reader immediately. The library refreshes as soon as each import finishes.

Aozora Bunko

Aozora Bunko (青空文庫) is a volunteer archive of public-domain Japanese literature.

The catalog

• Shiori fetches Aozora's full catalog CSV once, in the background, when the app starts. Per Aozora's politeness guidance, the download comes from the project's GitHub mirror, not aozora.gr.jp itself.
• The catalog is cached in the data directory (aozora_catalog.zip). If you are offline, the last cached copy is used; if there is no cached copy yet, the view shows "Catalog not available" until you reconnect.
• The ⟳ Reload catalog button next to the search box deletes the cache and fetches the current catalog.

Only importable entries are kept from the catalog:

Searching

Search is local against the cached catalog — instant and offline. Typing in the search box filters as you type; a result matches when the query is a substring of the work's title, title reading (kana), or author, so both 坊っちゃん and なつめ find their books. The first 50 matches are shown, each with the author and the work's orthography (e.g. 新字新仮名). With an empty query the view shows how many works are available.

Importing

Clicking ⬇ Import downloads the work's XHTML file — from the GitHub mirror first, falling back to aozora.gr.jp — and feeds it through the HTML import pipeline (furigana ruby markup is stripped). Aozora's servers send no charset header, so the file is decoded according to the catalog's encoding column: UTF-8 when the catalog says so, Shift_JIS otherwise. Title and author come from the catalog; the publisher field is set to 青空文庫.

Wikisource

The Wikisource tab searches Japanese Wikisource (ja.wikisource.org) — classic literature, historical documents, speeches, and law texts.

Searching

Unlike the Aozora tab, this is a remote query: type a query and press Enter or click Search. Shiori runs a MediaWiki full-text search over mainspace articles and shows up to 20 hits, each with its title, word count, and a text snippet around the match.

Importing

Clicking ⬇ Import fetches the page's rendered (Parsoid) HTML from the MediaWiki REST API, strips it to plain text, and imports it. The page title becomes the document title and the publisher field is set to Wikisource; there is no author metadata in the search results, so the author field is left empty (you can edit it afterwards in the library).

Politeness

Shiori is deliberately gentle with both services:

• Every request carries a descriptive User-Agent identifying the app, as the MediaWiki API policy requires.
• Requests run serially — catalog fetches, searches, and imports happen one at a time, never in parallel bursts.
• The bulk Aozora catalog comes from the GitHub mirror rather than the Aozora site, and individual work downloads also try the mirror first.
• Wikisource search requests pass maxlag=5, so they back off automatically when the servers are under load.

Offline behavior

After the first successful catalog fetch, Aozora search works fully offline. Importing a work and anything on the Wikisource tab require a connection. These catalog and import fetches, plus optional LLM calls, are the app's only network features — see Getting-Started for the first-run reference-data download.