Frequently Asked Questions

A structured kanji dictionary for Japanese learners who want to understand, not just memorize. Every character is broken down into its component graph (radicals, graphemes, kanji-within-kanji), every reading in every word is attributed back to a specific kanji and reading type, and an FSRS-5 spaced-repetition system layers on top so you actually retain what you look up.

Intermediate-to-advanced learners who are past the kana stage and want reference material that goes deeper than a general dictionary. If you've ever stared at 朝日 and wondered why 朝 is read あさ here but ちょう in 朝食, or what 廴 in 建 actually is, this is built for you.

The design is unapologetically dense — readings, radicals, frequency ranks, Heisig indices, stroke order, component graphs, pitch accent — laid out so the information is there when you want it, not buried behind taps.

Jisho is an excellent general lookup tool — the canonical "I encountered a word, what does it mean" dictionary. My Kanji is the deep-dive companion to that. The differences:

• Per-kanji reading attribution. On every word page, each kanji's hiragana contribution is colored by reading type (kunyomi, onyomi, nanori, jukujikun) and linked back to the canonical reading it derives from — so 雨 (あめ) → 春雨 (はるさめ) is shown as a rendaku variant of あめ, not a mystery.
• Component graph. Every kanji links to the radicals, graphemes, and sub-kanji it contains, and to every kanji that contains it. Walk the decomposition tree as far as you want.
• Built-in SRS. Add anything you look up to a study deck without leaving the page.

Use both. They don't overlap.

WaniKani is a curriculum — a fixed-order course that gates content behind levels. My Kanji is a reference tool — you look up any character at any time, in any order, and study what you actually want to study. There are no locked characters, no level-ups, no SRS-only-on-content-you-just-learned constraint.

If you're already doing WaniKani, My Kanji works well as the place you go when you want to understand a character more deeply than the WaniKani mnemonics, or look up something WaniKani hasn't taught you yet.

Anki is a generic spaced-repetition shell — you bring your own deck. My Kanji ships with a domain-specific deck (Jouyou kanji + 10,000+ compound words), pre-rendered stroke order, component graphs, pitch accent, and reading attribution already in place. You don't curate cards, you just study.

Under the hood, the scheduler is FSRS-5 with an SM-2 warmup blend (see "What is FSRS-5?" below) — closer to Anki's modern FSRS plugin than to its default settings.

Use the search bar at the top of any page. Four query types are supported:

• The character itself — 水, 朝, 漢
• Unicode codepoint in hex — 6c34 jumps straight to 水
• English meaning — water, morning, road
• Romaji reading — mizu, asa, michi. Long-vowel variants (koohii, kōhī, kohi) all match the same word, regardless of how you type it.

Single-CJK-character and hex queries redirect straight to the kanji's page; everything else returns a ranked list.

Yes. Words have their own dictionary at /words — every entry has a romaji-based slug (asahi, nihon, keizai), so direct URLs are predictable. Word search supports the same surface / reading / romaji / meaning queries kanji search does.

The color gradient represents your mastery for each kanji, derived from FSRS stability:

• Gray — not in your study deck yet
• Red — new or recently failed (low stability)
• Orange / yellow — learning (stability building)
• Green — mature (stability above ~21 days)

If you're not signed in, every tile is gray — mastery only exists inside an account.

On every word page, each kanji's hiragana contribution is colored by how that kanji is being read:

• Emerald — kunyomi (native Japanese reading; e.g. 水 → みず)
• Blue — onyomi (Sino-Japanese reading; e.g. 水 → スイ)
• Violet — nanori (reading used only in names)
• Amber — jukujikun (whole-word reading not derived from any single kanji; e.g. 大人 → おとな)

The same color scheme is used for the readings list on the kanji page, so the visual mapping is consistent everywhere.

A radical is one of the 214 traditional Kangxi classification units — the categories paper dictionaries use to file kanji. Each kanji has exactly one classification radical.

A grapheme is a graphical building block that appears inside kanji but isn't a standalone character or a Kangxi radical: 廴 ("stretching legs"), 宀 ("roof"), 艹 ("grass top"). Most decomposition systems collapse these into radicals; we keep them separate because they describe what you actually see in the character, which is often more useful than the classification radical for learning.

The component graph uses both: every kanji decomposes into a mix of sub-kanji, radicals, and graphemes.

Yes:

• /kanjis — all 2,136 Jouyou kanji, browsable and searchable
• /lessons/:grade — kanji grouped by Japanese elementary grade (1–6 + S)
• /radicals — all 214 Kangxi radicals, grouped by stroke count
• /graphemes — every grapheme that appears in the dataset
• /words — all compound words

The Atlas (/components/:grade) is a navigable graph of how kanji are built from each other. Click a kanji and it becomes the focal node; parents (kanji that contain it) appear above, children (its components) below. Walk the graph as far as you want.

Grade 1 is free; grades 2–6 + S are Pro.

A directed parent → child edge in the component graph: the parent kanji's glyph visibly contains the child as a structural piece. We tag each edge with a position label using the traditional system — hen (left), tsukuri (right), kanmuri (top), ashi (bottom), kamae (enclosing), tare (hanging), nyou (wrapping). So 漢 contains 氵 in the hen position and 莫 in the tsukuri position.

Many radicals have positional variant glyphs. 水 (radical #85) appears as 水 when it's a standalone or bottom element, but reduces to 氵 when it's on the left (hen position) — same radical, same meaning, different written form for spacing reasons.

The variant glyphs for each radical are listed on its detail page, with the position they're used in.

Add any kanji or word to your deck from its detail page (or via the persistent study bar). The scheduler decides when to show you each card. After answering, rate your recall from 1 (Again) to 4 (Easy) — the algorithm uses that rating plus your history to pick the next review interval.

There are no daily quotas, no required streaks, no penalties for skipping a day. The cards wait.

FSRS-5 (Free Spaced Repetition Scheduler, version 5) is an open-source spaced-repetition algorithm. It models each card with two parameters — stability (how long the memory will hold) and difficulty (how inherently hard the card is for you) — and schedules the next review at the moment your recall probability drops to a target threshold. We use 90% as the target.

Compared to SM-2 (Anki's default), FSRS reacts faster to lapses, is less punishing on Hard ratings, and produces more accurate intervals once it has data to learn from. The tradeoff is that FSRS needs reviews to calibrate.

FSRS is data-hungry — early in a card's life, the predicted intervals are noisy. So we run a hybrid scheduler: pure SM-2 below 30 reviews, a linear blend between 30 and 200, pure FSRS above 200. SM-2 state is always maintained alongside FSRS so the blend stays warm.

This is invisible to you — you just rate cards 1–4. But it explains why early cards feel stable and predictable, and why later cards adapt more sharply to your performance.

• 1 — Again. You couldn't recall it, or you got it wrong.
• 2 — Hard. You got it, but it took effort or felt uncertain.
• 3 — Good. You knew it.
• 4 — Easy. Trivial — you'd want to see this much less often.

Rate honestly. Rating "Easy" on cards you actually struggled with is the most common way to make the algorithm push reviews too far out and then forget them.

• Stability — how many days the memory is expected to hold at 90% recall probability. The number that matters most for tracking mastery.
• Difficulty — 0–1 scale, how inherently hard this card is for you given your rating history. High difficulty cards get shorter intervals.
• Easiness — SM-2 easiness factor (1.3–2.5+). Used during the warmup phase before FSRS takes over.
• Reviews — total times you've rated this card.
• Lapses — times you rated 1 (Again) on a card you previously knew.

• Learn mode introduces new kanji/words you haven't seen before. Each session is a 5-item, two-phase batch: introduce (forward order) then consolidate (reverse order, with a longer gap on item #1 for desirable difficulty). 10 interactions per session.
• Review mode drills cards that are already in your deck and due for review.
• Smart mode (the default entry point in the study bar) clears reviews first, then transitions to learn — so you never let due cards pile up while learning new ones.

Yes — two parallel SRS systems. Kanji cards live on /lessons, word cards on /lessons/word/levels. They share the same FSRS engine but track progress independently, because knowing the kanji 朝 doesn't mean you know the word 朝食.

Each kanji or word in your deck rotates through four card types:

• Meaning recall — "what does this mean?"
• Reading recall — "how is this read?"
• Kanji from meaning — "which kanji means X?"
• Kanji from reading — "which kanji is read X?"

Each type has its own SRS state, so getting "meaning" on 水 doesn't shortcut "reading" on 水. Distractors in multiple-choice are pulled from the same grade where possible to keep the difficulty calibrated.

Browser-native — we use the Web Speech API (speechSynthesis) with the local OS Japanese voices. No audio is streamed from a server; it's synthesized on your device. Quality varies by OS (macOS Kyoko / Otoya, iOS, Android, Windows Haruka) but local voices are preferred over remote network voices for reliability.

If pronunciation isn't working, see the speaker icon's troubleshooting hint, or try a different browser — Chrome's Web Speech support has had quirks historically.

Tokyo-dialect Japanese has a pitch accent — a single drop position per word where the pitch falls. We store this as an integer per word: 0 = heiban (no drop), 1 = atamadaka (drop after mora 1), n = drop after mora n. The conventional descending-line diagram is rendered on each word page.

Coverage: ~9,700 words have pitch accent data sourced from the kanjium dataset.

Pitch accent data isn't universally available, and different sources disagree. We only show it for words where we have a confident value. Multi-pattern words (where Tokyo dialect itself accepts more than one accent) are stored with a single canonical value.

Yes. Browsing the dictionary, searching kanji, viewing readings, components, words, radicals, graphemes, and Kanji of the Day are all free forever and don't require an account.

With a free account, you also get spaced-repetition study for Grade 1 kanji (the first 80 characters Japanese kids learn).

Pro unlocks SRS for grades 2–6 + S (the rest of the Jouyou set), and the Kanji Atlas component-graph navigator.

Three options:

• $7/month — billed monthly, cancel anytime
• $59/year — billed annually, works out to ~$4.92/month
• $199 lifetime — one payment, no renewal

Lifetime is capped at 1,000 sales total.

Any first-time Pro purchase (monthly, annual, or lifetime) is fully refundable within 3 days of the charge, no questions asked.

After 3 days: monthly and annual plans don't refund, but you can cancel anytime via /billing and access lapses at the end of the period you've already paid for. Lifetime is non-refundable after the 3-day window.

If something genuinely sympathetic comes up (medical, financial), email us and we'll work it out.

The lifetime of My Kanji as a service. If we shut down, lifetime access ends — there's no escrow. This is disclosed at the Stripe checkout itself.

If the service is sold or merged, the lifetime obligation transfers to the acquirer.

Yes. Email us and we'll handle the switch through Stripe — typically prorated against your current period.

/billing opens the Stripe-hosted Customer Portal. Click "Cancel plan." Your access continues until the end of the current billing period.

You can also delete your account entirely (see "How do I delete my account?" below) — that cancels any active subscription as part of the deletion.

Yes. Pro is a single tier; everything we add to it lands for all current Pro subscribers automatically. There won't be an "Pro Plus" upsell.

Seven: English, Japanese, Chinese (Simplified), Korean, Indonesian, Vietnamese, and Malay. Switch via the locale switcher in the navbar.

The locale prefix in URLs is optional — /kanjis/6c34 and /ja/kanjis/6c34 both work; the second pins Japanese for that visit.

Yes, where we have data. Each kanji, word, radical, and grapheme has a canonical English meaning and per-locale translations layered on top. Search is locale-aware: searching in Vietnamese runs against the Vietnamese meanings, in Korean against the Korean meanings, etc.

Coverage varies by locale. English is complete; other locales are partial and growing — entries without a translation fall back to English.

No, for browsing. Yes, for studying — SRS state is per-user, so it has to be tied to an account.

Yes. Click "Sign in with Google" on the login or signup page. We receive only your email and name from Google — nothing else. We don't post on your behalf, read your contacts, or touch any other Google data.

"Forgot password?" on the sign-in page sends you a reset link by email. The link is single-use and expires after 24 hours.

Not yet through a self-serve UI. If you need a CSV of your review history or card states, email us.

Email support@mykanji.app and we'll delete your account, study data, billing records, and email subscriptions within 30 days. Stripe-required transaction records are retained per their mandated period (typically 7 years), but those don't include any non-billing data.

A single SQLite database on a Hetzner server, with continuous Litestream backups. We don't sell, share, or use your study data for anything beyond running the service. Email is sent via a transactional provider; errors go to Sentry with a 10% sampling rate.

All 2,136 Jouyou kanji (the official set used in Japanese schools and newspapers). Each has full readings, meanings, components, stroke order, frequency rank, Heisig index, and grade.

We don't currently cover Jinmeiyo (additional name kanji) or Hyogai (rare/historical kanji). If a specific non-Jouyou character matters to you, let us know — coverage decisions are demand-driven.

Intentional. The grade system (Japanese elementary 1–6 + S for secondary school) is the only difficulty axis we model. JLPT level mappings are inconsistent across sources, change between editions of the test, and add a second axis that mostly duplicates grade. One axis, used everywhere — that's the design.

10,476 — chosen for frequency, pedagogical value, and good distribution across kanji. Each word has a surface form, full kana reading, romaji, meanings, per-kanji reading attribution, and (where available) pitch accent.

A mix of public datasets and manual curation:

• Kanji and component data from KANJIDIC2, KanjiVG (stroke order), and the Kangxi radical reference.
• Word data from JMdict (meanings, parts of speech), with curated reading attribution layered on top.
• Pitch accent from the kanjium dataset.
• Translations and pedagogical organization are our own.

All upstream data is used under the relevant licenses (EDRDG / CC).

Use the feedback form linked from the navbar, or email support@mykanji.app. Include what you were doing, what you expected, and what happened — screenshots help.

Please tell us. Linguistic edge cases (rendaku, lexicalization, regional variants, dictionary disagreements) are exactly the kind of thing that benefits from real users pointing out errors. Include a source if you have one — JMdict entry, dictionary citation, native speaker confirmation — and we'll fix it.

If it's outside the Jouyou set, see the "Which kanji are included?" question above. If it should be there, file a bug — that's a data import issue, not a coverage decision.

support@mykanji.app for everything — support, billing, data corrections, partnership inquiries, the lot.