Data & Interop

Where Shiori keeps your data, how to move cards to and from Anki, and how to back up or restore the database. Everything described here lives under Settings → Data unless noted otherwise.

Data directory

All app data lives in %APPDATA%\shiori (for example C:\Users\you\AppData\Roaming\shiori). The exact path is shown at the top of Settings → Data. If you used the app before its rename, the old japanese-reading-companion directory is renamed to shiori automatically on first launch.

Language packs

Removing a pack under Settings → Languages deletes its files under packs/<code>/ — nothing else. The language's library, vocabulary, and review history stay in the database and come back if the pack is reinstalled.

Anki export

Settings → Data → Anki → Export deck (.apkg) writes every SRS card — across all languages — as a note with four fields: Expression, Reading, Meaning, Sentence. The package uses Anki's legacy schema-11 format, which every Anki version can import.

• Scheduling is approximate. Shiori schedules with FSRS; Anki's classic scheduler is SM-2. At Anki's default 90% desired retention an SM-2 interval roughly equals FSRS stability, so the export maps stability → interval and difficulty → ease (difficulty 5 ⇄ ease 2500). New cards export as new.
• Re-exports update, not duplicate. Note guids are derived from the language and word identity (lemma, reading, part of speech), so importing a newer export into the same Anki collection updates the existing notes. Japanese words keep the historic jrc- guid prefix — exports made before 0.2.0 still update rather than duplicate — while other languages export under a shiori-<lang>- prefix.
• Exporting with no SRS cards is an error — learn some words first.

Anki import

Settings → Data → Anki → Import deck (.apkg) brings an existing deck into Shiori's SRS. For each note:

• The first field in the active language (HTML stripped) is morphologically analyzed, and its head token becomes the canonical word key — the same lemma + reading + part-of-speech identity used by reading and reviews. A deck imports into whichever language is active, so switch languages first if the deck isn't for the current one.
• SM-2 scheduling seeds the FSRS state approximately: stability ≈ the SM-2 interval (the 90%-retention equivalence again), difficulty from the ease factor, and the due date is preserved (clamped to 0–365 days out). Cards never reviewed in Anki import as new cards.
• Imported words with stability of 60 days or more are marked known; the rest are marked learning (see Reviews-and-SRS).
• Existing cards are never overwritten. A note whose word already has a card in Shiori is skipped, as are notes with no field in the active language. The result message reports imported and skipped counts.

One sharp edge: Anki's current default export uses a newer zstd-compressed format (collection.anki21b). Shiori rejects it with a clear message — re-export from Anki with "Support older Anki versions" checked.

Settings export and import

Settings → Data → Settings file. Export settings saves a copy of settings.json wherever you choose; Import settings loads such a file and applies it immediately. The file includes any LLM API keys you have entered under Settings → AI, so treat exports as secrets if you share machines. It also carries the active language, the per-language LLM model overrides, and the pack catalog URL, so those travel with an export.

Database backup and restore

Settings → Data → Database.

• Back up database writes a clean single-file copy via SQLite's VACUUM INTO. It is safe to run while the app is open — pending write-ahead-log contents are folded into the copy.
• Restore from backup does not touch the live database immediately. The chosen file is staged as jrc.sqlite3.restore-pending and swapped in on the next launch; restart the app to complete the restore. Your current database is kept aside as jrc.sqlite3.pre-restore, so a restore is reversible until the next one overwrites that file.
• Upgrading to 0.2.0 also backs up automatically. Opening a pre-0.2.0 database first writes a one-time clean copy jrc.sqlite3.v7-backup next to the database (via the same VACUUM INTO, and only if that file doesn't already exist) before the schema migration rebuilds the tables with language scoping.

Offline behavior

After the first run all reference data (dictionary, frequency list, kanji data, JLPT lists) lives locally in the database and data directory. Only a few things need the network afterwards:

• LLM calls to remote providers (Anthropic or a remote custom endpoint). Local backends via Ollama stay offline — see AI-and-Chat.
• Source search in Online-Sources: Wikisource search and catalog/book downloads. The Aozora catalog falls back to the last cached copy when offline.
• Language-pack downloads under Settings → Languages: installing a pack from a URL and the Build-from-Wiktionary source downloads. Both are optional; Wiktionary downloads cache in web-sources/ and resume over HTTP range requests, so an interrupted fetch picks up where it stopped.

Reading, importing, word lookup, reviews, and statistics all work offline. First-run setup can also proceed without the dictionary download — see Getting-Started.