Daily Occurrences

An Anki add-on that builds a daily occurrence dictionary from your immersion. It connects to a websocket that streams the Japanese text you're reading (and can also follow the ebooks you read in Hoshi Reader), parses each line with the Jiten reader API, and writes a per-day frequency list of every word that appeared.

It's built to feed the Priority Reorder add-on: point its output at Priority Reorder's _seen folder and the words you read each day automatically become available to that add-on's seen: prioritization. See Pointing it at Priority Reorder.

It runs quietly in the background whenever Anki is open, reconnecting on its own if the text source goes away and comes back. You can pause it at any time from the summary window.

Requirements

• A Jiten account and API key. Parsing runs on Jiten's servers, so each recorded line is sent to api.jiten.moe under your key.
• Something to read from: a websocket text source (a texthooker, clipboard inserter, OCR feed, or anything else that emits the sentences you're reading as websocket messages), and/or ebooks in Hoshi Reader Desktop — see Reading ebooks in Hoshi Reader.

Install

Install from AnkiWeb.

Setup

The only thing you have to configure is your Jiten API key. Everything else has a working default; leave it alone unless you have a specific reason not to.

1. Generate a Jiten API key: log in at jiten.moe, open your settings page, and find the API Key section.
2. In Anki, open Tools → Add-ons, select Daily Occurrences, and click Config.
3. Paste the key into jiten_api_key and save.

The add-on starts recording immediately, as soon as it finds something on the websocket. Settings apply as soon as you close the config window — no restart needed.

Two optional settings are worth knowing about, but neither is required:

• output_dir — set this if you want the dictionaries to feed Priority Reorder. See below.
• websocket_url — set this only if your text source isn't on the default ws://localhost:6677.

What it produces

Daily dictionaries are written under the add-on's user_files/occurrence-dicts/ folder (or wherever you point output_dir), one folder per day:

occurrence-dicts/
  2026-06-20/
    index.json
    term_meta_bank_1.json

Each YYYY-MM-DD folder is a complete Yomitan-format occurrence dictionary; words are stored against their occurrence counts. A day runs from your configured cutoff hour to the next (by default it follows Anki's "next day starts at" preference), so late-night sessions stay in one file.

Old day folders are deleted automatically after 60 days — see delete_after_days.

Pointing it at Priority Reorder

The primary intended use case for this add-on is to feed Priority Reorder's seen: search. That add-on reads daily occurrence dictionaries from a reserved _seen folder inside its own user_files, one subfolder per day named YYYY-MM-DD — exactly the layout this add-on writes. So rather than leaving the dictionaries in this add-on's folder, point output_dir straight at that _seen folder and every day's words land where Priority Reorder expects them.

To find the path, open Tools → Add-ons in Anki, select Priority Reorder, and click View Files. The target is the user_files\_seen subfolder of the folder that opens (it doesn't need to exist yet — it's created on the first write).

config.json is JSON, so on Windows you must either escape each backslash (\\) or use forward slashes:

"output_dir": "C:\\Users\\you\\AppData\\Roaming\\Anki2\\addons21\\<priority-reorder-id>\\user_files\\_seen"
// — or —
"output_dir": "C:/Users/you/AppData/Roaming/Anki2/addons21/<priority-reorder-id>/user_files/_seen"

Once set, seen:1 in Priority Reorder matches today's words, seen:7 the last week, and so on.

Sharing that folder is safe. A day folder that this add-on didn't write is never read, never overwritten, and never deleted — it's identified by the title inside its index.json. If something else already owns the folder for today's date, recording for that day is skipped rather than merged into it, and the summary window tells you so.

Reading ebooks in Hoshi Reader

Set hoshi_enabled to true and ebooks you read in Hoshi Reader Desktop are recorded alongside your websocket text, into the same daily dictionaries. Reading done in Hoshi on your phone counts too, once Hoshi Desktop has synced it.

Hoshi keeps each book's EPUB, your current position, and (if you let it) your reading sessions in its library folder. Every hoshi_poll_minutes the add-on looks at how far each book has moved since it last looked, cuts that stretch of text out of the EPUB, and records its words — into the day you read them, which for synced phone reading may be an earlier day.

• Turn on statistics autostart in Hoshi's settings. Hoshi counts a jump through the table of contents as "characters read", so the add-on uses each session's reading time to tell reading from skipping: movement faster than hoshi_max_chars_per_minute while stats were running is treated as a skip and not recorded.
• Forgetting the stats costs you nothing. Movement with no session behind it is assumed to have been read and is recorded in full, to the day of your last page turn.
• Nothing is recorded twice. Each stretch of a book is recorded once, ever: going back, re-reading, re-importing the book or Hoshi syncing old sessions later never adds the same text again.
• Books you'd already read aren't backfilled. When you turn this on, recording starts from wherever each book is at that moment.
• Synced books need downloading. A book you've only read on another device is waiting in the queue until Hoshi Desktop downloads it — open it there once.
• Big backlogs trickle in. A day of phone reading is sent to Jiten a few thousand characters at a time, a few seconds apart, and at most 40,000 characters a minute. The summary window shows what's still queued.
• Run it in one program only. If another program is also reading your Hoshi library into the same folder, the add-on notices and stays out of the way rather than count everything twice.

Set websocket_url to "" if Hoshi is your only source.

The summary window

Tools → Daily Occurrences Summary opens a small panel showing:

• a status dot — solid green when it's connected to the websocket, solid red when it isn't, and split diagonally green/red while recording is paused,
• the date it's recording for,
• how many distinct terms and total terms you've recorded today,
• with Hoshi enabled, how many books it's watching, how much ebook text it has recorded today, and anything still queued or waiting for a download,
• a Pause recording button,
• a red line describing the problem, if anything is stopping words from being recorded — a missing or rejected API key, an unreachable server, or a day folder in output_dir that belongs to something else. If words aren't appearing, this is the first place to look; the full detail goes to user_files/daily-occurrences.log.

Pausing

Pause when you're about to read something you don't want in today's dictionary — re-reading a passage, testing your text hook, or just handing the machine to someone else.

While paused, text arriving on the websocket is thrown away as it comes in. Nothing is saved up and recorded later, so anything that arrives while paused simply never happened as far as the dictionary is concerned. The connection itself stays open, so resuming is instant.

Pausing doesn't affect Hoshi Reader. Ebook reading is still recorded while paused — including reading from your phone that happens to sync in while you're paused.

Text you read before pausing is still saved, so the counts may keep climbing for a few seconds after you click Pause before they settle.

A pause isn't remembered across restarts. Closing your profile or quitting Anki always resumes recording, so a pause you forget about can't cost you more than the one session. It does survive saving the config screen.

Settings

You (probably) should not need to change any of these, and I'd suggest not touching them unless you genuinely know what a given value does.

The defaults aren't the product of careful measurement — they're values I picked because they're conservative. Parsing happens on Jiten's servers, and a text source can misbehave in ways that have nothing to do with how much you're actually reading: a badly configured texthooker, a page that dumps an entire backlog at once. The batching, dedup, and line-length limits exist so that garbage like that gets dropped locally instead of turning into a flood of pointless requests against Jiten.

Every setting is also documented inline on the config screen. A value that isn't a number, or is below the minimum a setting allows, is replaced with a sane one rather than accepted — the substitution is noted in user_files/daily-occurrences.log.

jiten_api_key

Required. Your Jiten API key. Empty means recording is disabled.

websocket_url

The websocket to read text from. Default ws://localhost:6677. Point this at whatever your text source exposes. "" turns the websocket off.

output_dir

Where daily dictionary folders are written. Empty (the default) uses user_files/occurrence-dicts/ inside this add-on's folder. Set an absolute path to write elsewhere — typically Priority Reorder's _seen folder, see Pointing it at Priority Reorder.

delete_after_days

Delete a day's dictionary once it's more than this many days old, so the folder doesn't grow forever. Default 60. 0 = keep everything.

Only folders named YYYY-MM-DD that this add-on wrote are ever removed, so it's safe to point output_dir at a folder you share with something else. Keep this at or above the largest seen:N you actually search with in Priority Reorder — a day that's been deleted can't be matched.

jiten_base_url

Jiten API base URL. Leave as https://api.jiten.moe unless you self-host.

jiten_timeout_ms

How long to wait for a Jiten response before giving up on that batch, in milliseconds. Default 10000.

flush_every_lines

Process after this many new lines — also the batch size sent per API request. 0 = only process on the idle timer below. Default 50. (Regardless of this value, a batch is capped at ~4000 characters per request.)

Lowering it means more, smaller requests for the same amount of reading.

idle_flush_seconds

Process after this many seconds with no new line. 0 = disable. Default 30. This is what gets the tail of a reading session recorded once you stop.

Saving to disk is separate from either of these, and happens at most every few seconds while there's something new — so setting both to 0 still won't lose your day's words.

dedupe_window_lines

Drop exact-repeat lines seen within this rolling window before parsing, so re-sent clipboard/OCR frames aren't counted twice or sent to the API twice. 0 = disable. Default 500.

max_line_length

Ignore any incoming line longer than this many characters, before it's deduped or parsed. This is the main guard against a bad hook or a backlog dump flooding the dictionary with text you never actually read. 0 = no limit. Default 150.

day_cutoff_hour

The hour (0–23) at which a new "day" starts, so a late-night session stays in one dictionary. null (the default) follows your Anki "next day starts at" preference. Set a number 0–23 to override it.

hoshi_enabled

true to record ebooks read in Hoshi Reader. Default false. See Reading ebooks in Hoshi Reader.

hoshi_books_dir

Hoshi's library folder. Empty (the default) finds it automatically: %APPDATA%\hoshi-reader\Books on Windows, ~/Library/Application Support/de.manhhao.hoshi/Books on macOS.

hoshi_poll_minutes

How often to check Hoshi, in minutes. Default 1, minimum 1.

hoshi_max_chars_per_minute

The fastest reading speed the add-on will believe, in characters per minute. Default 1000, minimum 100. Movement faster than this while Hoshi's stats were running is treated as skipping ahead. It's a ceiling rather than an estimate, so it only needs raising if you genuinely read faster.

Third-party code

The websocket client under vendor/websocket/ is websocket-client, bundled unmodified under the Apache License 2.0 (see vendor/websocket/LICENSE).