There are plenty of Japanese dictionaries already. How is Tomoshi different?

In one line: most dictionaries are built for reading — looking a word up so you can get through the sentence in front of you. Tomoshi is built for producing — saying it and writing it correctly yourself.

Reading-oriented tools will get you through a text or an exam. But once you start trying to speak Japanese, or write an email that sounds right, you need a different set of things: which of a transitive/intransitive pair to use, how near-synonyms actually differ, which kanji spelling of a reading is the usual one, how spoken and written registers line up, fast offline lookup on the desktop, English-to-Japanese reverse lookup, and AI explanations that fit the sentence you are actually writing. Those are Tomoshi’s core features.

Is Tomoshi available in English?

Yes, though not every corner of it yet.

  • The interface has an English version — Settings → Interface language.

  • The dictionary data has an English pack: definitions, the orthography, nuance and transitive/intransitive notes, and English-to-Japanese reverse lookup all come in English. You choose your dictionary language on first launch and can change it later in Settings.

  • AI explanations come back in English when the interface is set to English.

  • The privacy policy and terms of service have English versions. They are translations provided for convenience — the Simplified Chinese versions are the originals and the ones that govern.

  • The reading cards — the title, synopsis and reading notes for each literary work — are in English too, written against the Chinese originals and reviewed by a separate model rather than machine-translated. The Japanese texts themselves are, of course, the originals.

What level is Tomoshi for? Can a complete beginner use it?

Tomoshi suits learners who already have a foundation — roughly N3–N2 and up — and are working on speaking and writing: people living or working in Japan, taking one-to-one lessons, or writing Japanese email and documents day to day. Using a transitive verb where its intransitive twin belongs, picking the wrong member of a synonym set, dropping written-register words into speech: those are the corrections that dominate the “can pass the test, can’t open my mouth” stage, and they are exactly what Tomoshi is built around.

A beginner can of course use it as an ordinary dictionary. But if your goal right now is to pass the JLPT, a mature exam-oriented dictionary may serve you better — come back when you start wanting to say it.

Is Tomoshi free? Will it start charging?

Looking words up is free forever, and that is not going to change: the complete dictionary data, offline lookup, select-to-look-up in the browser and the error-reporting feature are free now and will stay free.

Cloud features — AI explanations and cloud speech — are free to use today, within a daily quota. Every call to them has a real server cost behind it, which I currently cover myself; if the user base grows to the point where I can no longer cover the server and AI costs, they may become paid to balance those costs, and this would be announced in the app in advance. Put simply: the dictionary is free; the cloud services would only become paid if their costs became more than I can carry. I would rather state the rule from day one.

Which platforms does Tomoshi support? Is there a mobile version?

Tomoshi runs on Windows, macOS, iOS and Android: the desktop app is available to download now, the iPhone and iPad app is on the App Store, and the Android app is on Google Play.

Desktop came first because it overlaps with where output-focused learners actually work: looking words up while reading news or fiction on a computer, checking usage while writing an email or a document, searching quickly during an online lesson. In those situations, fast offline lookup on the desktop is something a phone app cannot match.

Does Tomoshi need an internet connection?

The core is fully offline: dictionary lookup, transitive/intransitive pairs, nuance notes and browser select-to-look-up all work with no network at all.

Two things do need a connection: AI explanations (which call a large language model) and read-aloud audio (cloud speech synthesis). Those are unavailable while you are offline, and work again as soon as you reconnect.

AI is so capable now — why not just ask a chatbot instead of using a dictionary?

This is the question I get most often. My answer is that the two are complementary, and that Tomoshi has AI built into it — the point is using AI where it is actually good.

Four differences. First, looking words up is a high-frequency, fragmentary act that may happen dozens of times a day; a dictionary opens offline and answers instantly, whereas opening a chat window and waiting for generation adds friction that accumulates badly. Second, a dictionary gives you structured, checked data: transitive/intransitive pairing and nuance notes are fixed and verified in Tomoshi, whereas a model may answer differently each time and can be confidently wrong — for language you are about to say out loud, that uncertainty is fatal. Third, what AI genuinely excels at is contextual explanation, so in Tomoshi the AI sits as a layer on top of the dictionary: the dependable skeleton comes from the data, and the explanation of the sentence in front of you comes from the model.

The fourth point is about learning itself. A language is not something you master by asking once; you have to meet it again and again until it sticks. In a chat app a good answer is quickly buried under everything else you asked that week, and finding it again — let alone reviewing it — is painful. I asked myself the same Japanese question many times over before it finally stuck. In Tomoshi an AI answer can be saved to your word list in one click and enters the same daily review flow as vocabulary; and when you search, the box surfaces AI results you have looked at before. No more asking the same question five times and getting five different answers.

Where does the dictionary data come from? Can it be trusted?

Tomoshi’s data comes in two layers. The base definitions come from established open dictionary projects such as JMdict, which the global Japanese-learning community has been checking for years. The production-oriented data that sets Tomoshi apart — transitive/intransitive pairing, nuance notes, kanji spelling comparisons, English-to-Japanese reverse lookup — I built with AI assistance and cross-checked across several models. Pronunciation data for common words has a separate quality line: it is checked by hand, working down by frequency.

A dictionary should have zero tolerance for errors, so Tomoshi has one-click error reporting built in, and every error you report is dealt with seriously. I think an open, traceable correction process deserves more trust than a claim of having no errors.

Is my search history uploaded anywhere?

No. Local lookups happen entirely on your device, and the history stays there — nothing is sent to a server.

Only when you deliberately use an AI explanation or cloud speech does that particular word or sentence get relayed through Tomoshi’s server to the relevant provider, in order to produce the result. Nothing else is collected, and using Tomoshi requires no account.

Is Tomoshi open source?

Partly. The browser select-to-look-up extension is open source under AGPL-3.0 — it has to read the pages you browse, which makes it the most privacy-sensitive piece, so that part is code you can read for yourself. The parts of the dictionary data that derive from open projects such as JMdict, along with content derived from them, are also available for download under CC BY-SA.

The desktop client is not open source yet: getting a codebase into publishable shape is a substantial amount of extra work for one person, and for now I would rather put that effort into the product itself. Opening it up is planned for a later version.

On privacy you do not have to take my word for it: Tomoshi only goes online when you deliberately use an AI explanation or cloud speech, and you can verify that directly with a firewall or a packet capture.

Can I sync my word list and history across devices?

The wordbook, yes. Search history, no — and that is deliberate.

Sign in with an email verification code on any device (“Me / Settings → Account”) and your wordbook and review progress stay in sync across every device you sign in on: once after launch, a few seconds after each change, and whenever the app returns to the foreground. Without signing in nothing changes — everything stays on that device.

Search history never goes to the cloud. It records what you looked up, which is far more sensitive than a word list, and the convenience of syncing it is not worth that exposure. Sync covers the wordbook table alone, and the privacy policy says so explicitly.

Sync is free for now. It is one of the “cloud features with real server costs” mentioned above and would only become paid, together with the other cloud features, if the server costs became more than I can carry; you will be told in advance inside the app, and the wordbook on each of your devices, the export feature and everything already synced to them will keep working.

Can I export my word list to Anki?

Yes. “Export to Anki”, next to “Sync now” on the wordbook page (also under Me / Settings → Account → Wordbook data), writes a CSV file Anki imports directly. The word goes on the front; the back is as complete as the app’s review card: the reading first, then every meaning of a word with its part-of-speech and usage tags and its own example sentences; a grammar point’s meaning, formation, usage notes, examples and the contrast with easily confused points; a saved AI analysis laid out by type and marked as AI-generated at the end. Choose the Basic note type when importing — the import dialog preselects whichever note type you used last, which may not be Basic. Separate headword, reading, furigana and meaning fields plus JLPT-level and mastery tags come along for custom note types. The file carries Anki’s note-identifier column, so exporting again updates the same notes instead of duplicating them.

To import: on desktop Anki and AnkiDroid choose the file from Import; on iPhone save the file to the Files app, open it and choose AnkiMobile.

Review progress is not included yet; an .apkg export that carries it is planned. There is no import in the other direction: every word in the list must map to a dictionary entry, and matching Anki’s text back to entries cannot be done reliably.

How are review intervals scheduled?

The word list schedules reviews with FSRS, the same open-source algorithm built into Anki. From each word’s memory stability and difficulty it estimates when you are about to forget it and places the review just before that point, so the same amount of reviewing holds more. You still just answer “Forgot” / “Knew it”; existing progress carries over without a reset.

The word list is a graduation funnel: once a word’s interval passes two months and you still recognise it when it comes due, it leaves the queue for “Mastered” and can be recalled any time. The daily review goal and limit are adjustable under “Me / Settings → Account”.

How do I delete my account and my data?

  • Delete your account: “Me / Settings → Account → Delete account”. Deletion takes effect immediately: the server removes your sign-in identifier (email address), account record and wordbook cloud copy, and this cannot be undone. The wordbook and history on the device itself are unaffected — clear them separately if you wish.
  • Local data: history and wordbook entries can be cleared one by one or all at once inside the app; uninstalling the app deletes all local data.
  • Everything else: data not tied to an account, such as feedback and crash reports, is deleted automatically at the end of the retention periods stated in the privacy policy; questions go to [email protected].

I found a wrong definition, or I have a suggestion. How do I report it?

For data errors, the easiest route is the one-click error reporting inside the app. For feature suggestions and anything else, email me at [email protected]. As a solo developer I read every message myself — and user reports are genuinely a working part of how Tomoshi’s data quality is maintained.