Privacy policy

Natsukashi is a personal Japanese vocabulary-mining and spaced-repetition app. This page describes, in plain language, what the app stores, what it sends to outside services, and why — based on what the app's own code actually does. We'll update this page if anything changes.

Your cards, review history, Known words, settings, AI credit counts, any feedback text you send, and error reports are stored in Firebase (Google Cloud, in London, europe-west2) — specifically Firebase Authentication (to identify your account) and Firestore (a database) for everything else.

If you use Natsukashi without signing in, you get an anonymous Firebase account tied to this device/browser, not to any personal detail like an email address. If you sign in with Google or email, your cards and progress move to that account so you can use them on another device. For email sign-in, Firebase Authentication stores your email address and a securely hashed password; Natsukashi never sees your password.

Natsukashi uses a small number of AI providers to power specific features, each only when you use that feature:

• Google Gemini (gemini-3.8-flash) receives: the Japanese text or screenshot you submit under Add Japanese (to find candidate vocabulary and write an example sentence); the rows you ask Fill with AI to complete during import; a screenshot when you use on-screen text extraction (OCR); the word/sentence details used to write an image prompt when you generate a card image; the word or phrase you look up in the Dictionary; and the target word when you ask the Dictionary to generate graded example sentences.
• Gemini's text-to-speech model (gemini-3.1-flash-tts-preview) receives the Japanese text or reading for a card to make the natural voice, either when it is first played or ahead of time for the cards due in today's review. The clip is then saved on the server (with only that text, not who asked for it) so the same text is never sent twice. Offline, your device's own voice may read the text instead; it stays on your device.
• OpenAI (gpt-image-1.5) receives the AI-written image prompt when you generate a flashcard image, and returns the generated image.
• Wikimedia Commons receives your search text when you search for a photo to attach to a card, and serves the image itself (Natsukashi's server relays the image so your browser never contacts Wikimedia directly).

None of these requests include your name, email, or any other card's content — only what the specific feature needs to do its job.

Natsukashi shows no ads and uses no third-party analytics or tracking of any kind.

During the private beta, when you are signed in, the app records a few facts about how it is used: when you open it (iPhone, Android or computer; installed or in the browser), when you finish a stretch of reviewing (how many cards, how many Again, how many minutes), which Dictionary tab you search in (not what you search for), when you tap a speaker, when you finish the level step or add a Starter Collection, which style an image was made in and whether it worked, and for each AI request which feature it was, whether it worked and how long it took, plus when a limit is reached.

The app's owner sees these together with counts from your account (when cards were added and how, review dates and ratings, AI credits used, error reports), next to your account email, only to improve the app during the beta. It never includes the Japanese words, sentences, search words or images in your cards. There are no cookies or advertising IDs, and someone using the app without an account sends nothing. This usage data is kept until the end of the beta (12 months at most) and is deleted with your account.

If something breaks, the app sends a short technical error report (the error message, the screen it happened on, the app version and your browser type) so it can be fixed. It is linked to your account id, not your name, and is deleted with your account.

In the app, go to Profile → Delete account and data. This immediately and irreversibly deletes your cards, review history, Known words, settings, feedback, error reports and usage data, both in your account and on this device. Backups of the underlying database are kept for a short window (currently up to 14 days) before they are automatically deleted — see this project's backup runbook for details.

Natsukashi is run by Chris Angelidakis. Questions or requests about your data can be sent to angelidakis.chris@gmail.com, or through Send feedback in Profile.