Your data
Privacy notice
Effective July 13, 2026 · Revised November 20, 2026
This notice explains what Kotomata processes during its invitation-only alpha, what normally stays on your device, and the choices available to you.
Account identity
Cloudflare processes your account identifier, email address, authentication session, and related security metadata so you can open Kotomata, securely use account-gated AI features, and synchronize supported data. If you choose Continue with Google, Google confirms your identity. Resend delivers account email. Kotomata does not sell personal information.
Each signed-in session also keeps a short label for its browser, such as “Chrome on Windows” (made from what the browser reports about itself at sign-in, not the full report), and the country Cloudflare reports for its connection, updated at most once a day. They are only for Signed-in devices under You → Account, where you can sign a device out, and are deleted when that session ends.
To stop abuse, Kotomata's server counts sign-in attempts by IP address and by email address. It doesn't store either: each count is kept under a one-way code made from the address, which can't be turned back into it, and is deleted after 24 hours.
Kotomata sets two cookies, both needed to sign in, and no tracking cookies: sid keeps you signed in (60 days, renewed while you keep using Kotomata), and g_state protects Continue with Google while you sign in (10 minutes).
Local study data and cloud sync
Your SRS history, settings, imported learning state, and other study data are primarily stored in your browser using local storage and IndexedDB. While your account session is active, selected data is synchronized to Kotomata’s server on Cloudflare so it can be restored or used across devices. Integration secrets are excluded from ordinary sync.
Kotomata also keeps a record of each review answer in your account: which item, how and when you answered, and whether it counted. Besides scheduling your reviews, Kotomata uses these records to check and tune how its review schedule works for everyone. For that, the records of many learners are analysed together under a code in place of your account, without your email, item names you created, or any text you wrote or imported. The combined copy is made for each analysis and not kept. You can opt out in Settings, and the records are deleted with your account.
This local-first design lets already-downloaded study features keep working during connectivity loss. It does not create an anonymous mode: signing out locks the application on that device while retaining its account-owned cache for a later sign-in.
Imports and generated content
Books, transcripts, images, PDFs, audio, URLs, annotations, and generated study material may be stored locally. Content is sent to a remote service only when you choose a feature that requires upload, cloud storage, transcription, analysis, synchronization, or AI processing. Avoid importing content you lack permission to use.
Imported personal material and your voice recordings can be especially sensitive. They stay on the device during local use, but leave the device when you invoke remote AI, transcription, cloud sync, sharing, or another clearly networked feature.
Books kept in your account. When you import a PDF, EPUB, or text file while signed in, a switch (on unless you turn it off) keeps the book in your account so it opens on your other devices. For a PDF, Kotomata stores the original file; for every book, it stores the processed text and word list. These copies are kept in private, per-account Cloudflare R2 storage, encrypted at rest but not end-to-end, and are never shared, made public, or deduplicated with another user's copy. Limits are 100 MB per file and 2 GB of books per account; a book past them stays on your device only. Deleting the book, or your account, deletes the stored copy. Built-in sample books are not stored, since they download again from the site.
Mined cards and private media
When you explicitly mine a sentence or subtitle, Kotomata stores its text, source position, and linked vocabulary as private study data. The YouTube extension can also capture up to 15 seconds of audio rendered by your browser and one still image, but only after you separately enable media capture in Watch-along. If capture is unavailable, the card is saved without media.
The extension temporarily keeps unacknowledged captures on your device until the same signed-in Kotomata account durably imports them. Kotomata then stores media locally for playback and transfers it directly to private, per-account Cloudflare R2 storage using a short-lived authorized URL. Kotomata’s server authenticates the request and records account-level usage, but the media is not public, searchable, shared, or deduplicated with another user's copy.
AI processing and your own API key
When invoked, chat prompts, relevant excerpts, responses, generated content, audio transcription, and speech go through Kotomata's server and OpenRouter to the company behind the model in use (listed under Service providers). Kotomata picks the model for every task, chat included; you don't choose one. Your OpenRouter key is sent to Kotomata's server, stored encrypted, and never returned to the browser. Provider credentials are not part of cloud study-data sync. Because the key is yours, your OpenRouter account's settings decide whether OpenRouter and the model's company keep your requests.
For each AI request, Kotomata logs its task, model, size, cost and outcome with your account, never its text, for 90 days, then keeps only monthly totals. When you add AI readings to a YouTube video's captions, Kotomata keeps the readings it gets back, as positions in the caption text, and may show them to other learners watching the same video, never with your name or email. Your own readings are deleted with your account.
Integrations and browser extension
WaniKani and JPDB credentials are cached on the signed-in device because the browser calls those services directly, and can also be stored encrypted through Kotomata's credential service for use on your devices. AnkiConnect is contacted locally on your computer. Integration metadata such as enabled service names, sync times, item identifiers, and failures may be stored for sync and diagnostics.
Videos play in YouTube's own player (youtube.com) and show YouTube's thumbnails, so Google receives your IP address and may set cookies under its own privacy policy. When you look up a book by its ISBN, your browser asks OpenLibrary directly; when you open or add a Syosetu web novel in Kotomata, it asks Syosetu's public API for the novel's details directly. Each of them sees your IP address.
The browser extension runs on its own on two sites only: YouTube (www.youtube.com), where watch-along starts by default on each video, and Syosetu (ncode.syosetu.com), where it reads and scores each novel chapter you open, without a click. Watch-along holds the video while it reads and scores the captions, then shows them in its caption shelf; closing the shelf turns it off for every video, and videos are then still scored in the background. On other sites it reads the active page's text only after you press Score or Open in reader. It runs no script on isori.space: the Kotomata app there talks to it through the browser to check that it is installed, pair your sign-in, and collect cards you mine. Scoring and dictionary lookup are local. If you sign in, an extension session, compact vocabulary snapshot, private resume pointers, explicit word actions, and mined-card handoffs can use Kotomata’s server on Cloudflare or the private media flow described above. For each web page or video you open in the extension's reader, watch along with, or add to your Library from the extension, it also saves the page's word counts to your account, privately: which dictionary words it uses and how often, never its text, so History and the Library can show how much of it you know. For a web page you track in your Library, it also saves the set of words in each sentence (sentences in page order, words in no order), where each screen-sized part of the page starts, and which grammar patterns the page uses, so the page's Library card can show how ready you are for each part of it. These are never shared or published, and Forget page… removes them with the page. When watch-along reads a YouTube video's captions, the extension also saves those caption lines and their timings to your account, privately, for every video you watch along, with no step from you and no setting to turn it off, so Kotomata's own player can play the video with them on another device. They are never shared or published. They are kept while the video is in your History or Library and removed with it: choose Forget page… in the video's menu in History to remove one. Community catalog contribution covers watched YouTube videos only, never web pages. It is on unless you turn it off in the extension's popup, and it skips videos YouTube marks unlisted, private or age-restricted and videos with little Japanese in their captions. It sends the video's link, title, unordered vocabulary counts, and facts about the video's language (the audio language YouTube lists, the language of its auto-generated captions, how much of the video has Japanese captions, and whether they come in long stretches or short bursts), not page or transcript text. Turning it off stops new contributions and cancels any still waiting, and the extension remembers that choice on the device even after you sign out.
Chrome Web Store Limited Use
Kotomata Companion uses information handled through browser permissions only to provide or improve its single purpose: helping you assess and study Japanese content using your Kotomata learning state. It does not use or transfer that information for personalized advertising, retargeting, creditworthiness, or sale to data brokers.
Extension data is transferred only when needed to provide a feature you choose, to maintain security, to comply with law, or as part of a merger or acquisition subject to this notice. No person reads page text, captions, authentication data, or private study data except with your specific consent for support, when necessary for security, when required by law, or after the data has been aggregated and anonymized for internal operations.
Service providers
- Cloudflare delivers the app and dictionary files, runs sign-in, your account, synced study data and Kotomata’s server, privately stores the media of cards you mine, counts page views without cookies (Web Analytics), and checks that a request for a sign-in email comes from a person, not a bot (Turnstile, which loads with the sign-in form and reads signals from your browser).
- Google confirms who you are when you choose Continue with Google, serves the app’s typefaces (Google Fonts) once you are signed in, so your browser’s IP address reaches Google then (the public pages and the home page’s demo use Kotomata’s own copies), and plays videos in YouTube’s own player, which sees your IP address and may set cookies.
- Resend sends sign-in and account email.
- OpenRouter passes AI requests, using your own key, to the company behind the model in use: Anthropic (Claude Haiku) for most grading, sentence building, grammar review and page reading; OpenAI (GPT) for chat, the other AI tasks, caption readings and transcription; and a host of the open Kokoro voice model for spoken audio. Kotomata picks every model.
- Sentry receives error reports, filtered so they carry no credentials, email addresses or study content, and a short record of each app session (release, browser, and your account ID once you are signed in), in the United States.
- OpenLibrary answers book lookups by ISBN and serves book covers, straight from your browser.
- Syosetu answers lookups of a web novel’s details when you open or add one in Kotomata, straight from your browser.
Payments are off during the alpha.
Error reports, visit counts, and diagnostics
Sentry is in the United States and keeps error reports for 30 days. When Sentry is enabled, Kotomata sends technical error, release, environment, browser, and coarse device information after filtering sensitive fields. Reports are designed not to contain credentials, email addresses, prompts, transcripts, imported text, audio, or study answers. Diagnostics copied by you are previewed and contain configuration state rather than study content or tokens.
Each time the app opens, Sentry also receives a short session record, even when nothing goes wrong: the release, environment, and browser, whether the session hit an error, and, once you are signed in, your account ID. It holds no page text or study data. Kotomata uses it to see how often each release fails. As with any request, your browser's IP address reaches Sentry with it. Error reports and session records carry your account ID once you are signed in, never your email.
Kotomata's server on Cloudflare doesn't keep request logs.
Cloudflare Web Analytics counts page views on isori.space, without cookies. For each page you open, it records the page's address, the site that linked to it, your browser, device type and country, and how fast the page loaded. Kotomata sees these as totals, not as a record of any one visitor.
Kotomata counts visits that arrive from AI-assistant links (such as a ChatGPT answer), by assistant name and day. To count each visitor once a day, it keeps a one-way code made from your IP address (on IPv6, its network part) and a random value used for that UTC day only. Both are deleted early the next morning (UTC), after which the code can't be matched to an address. No account, cookie, or page is recorded, and the daily counts are deleted after 30 days.
Where Kotomata lists shops and libraries for a book (Where to get it in Discover, Get it on a Library card, the store link under the textbook you pick during setup, the next volume's store link on a textbook in your Library, and the Read the original link on a film in Discover), it counts how often each title's list is opened and how often each link in it is followed, by title, store, where the link was shown, and day only: no account, cookie, or address is recorded with the count, and the link opens the store directly, not through Kotomata. A store doesn't tell Kotomata what you buy. The daily counts are deleted after 400 days. To pick which store those links open, Kotomata's server reads the country Cloudflare reports for your connection each time the app checks your sign-in; the app keeps it on your device, and the server doesn't store it.
Retention, export, and deletion
Device data remains until you clear it, the browser removes it, or you uninstall the app. Cloud account data remains while the account exists and for limited backup, security, or legal periods after deletion. Before Kotomata's server moved to Cloudflare, it ran on Supabase (United States). That earlier server is paused, receives no new data, and will be deleted; until then it may still hold account and study data saved before the move. Private media that is no longer referenced by any synced mined card may be removed after a 30-day cleanup grace period. Caption lines saved from videos you watched along are kept while the video is in your History or Library (in History, only the 300 most recently watched videos keep theirs) and are removed with the video or with the account. You can export study data from Data Management, including a mined-card bundle containing locally present media. Clearing device data is separate from permanently deleting an account. Account deletion removes account-linked database records and private mined media from Cloudflare R2, plus public contributions where safe; non-identifying shared catalog facts already incorporated into a canonical work or dictionary result may remain without your identity or private source material. Other devices keep their offline copies until this site's data is cleared in that browser, or another account signs in there.
Your choices, access, and correction
You can turn Community contribution off, sign out of or uninstall the extension, export your study data, clear a device copy, and delete your cloud account. You may also contact support@isori.space to ask what account-linked personal information Kotomata holds, request a correction, withdraw consent for optional processing, or raise a privacy complaint; Kotomata replies within 30 days. If you are not satisfied with the answer, you can complain to the Office of the Privacy Commissioner of Canada, to the Commission d'accès à l'information du Québec, or to the privacy regulator where you live. Kotomata cannot provide the application when processing required for authentication and account-gated services is unavailable.
Payments, security, and international processing
If payments are later enabled, Stripe would process payment and billing details; Kotomata would not store full card numbers. Kotomata uses access controls, encryption in transit, and scoped server credentials, but no internet service can promise absolute security. The service providers listed above may process and store data in the United States and in other countries outside Canada, where local authorities may be able to lawfully access it.
Who is responsible, and contact
Kotomata is operated by Buretto Technologies Inc., a company in British Columbia, Canada, which is the controller of the personal data described above. Its mailing address is 422 Richards St, Suite 170, Vancouver, BC V6B 2Z4, Canada. The person in charge of protecting personal information is the Founder of Buretto Technologies Inc., reachable at support@isori.space. Contact support@isori.space with privacy questions, access or export requests, or deletion problems. Material changes will update the revision date.