Metadata-Version: 2.4
Name: vocabbot
Version: 1.0.0rc11
Summary: Vocabulary in one to three languages at once with spaced repetition, in the terminal and in Claude or Codex.
Keywords: vocabulary,spaced-repetition,fsrs,language-learning,claude,codex
License-Expression: Apache-2.0
License-File: LICENSE
License-File: NOTICE
Classifier: Development Status :: 4 - Beta
Classifier: Environment :: Console
Classifier: Intended Audience :: End Users/Desktop
Classifier: Natural Language :: Polish
Classifier: Operating System :: OS Independent
Classifier: Programming Language :: Python :: 3
Classifier: Topic :: Education
Requires-Dist: mcp>=2.3.0,<3 ; extra == 'server'
Requires-Dist: tzdata>=2026.6 ; extra == 'server'
Requires-Python: >=3.10
Provides-Extra: server
Description-Content-Type: text/markdown

<h1 align="center"><img src="https://raw.githubusercontent.com/dithiothreitol/vocabbot/main/docs/assets/logo.png" alt="VocabBot" width="480"></h1>

**English** · [Polski](README.pl.md)

VocabBot teaches vocabulary in one, two or three languages at once: English,
German, Spanish, French and Italian, from A1 to B2, for learners whose base language
is Polish. It follows the way the SuperMemo vocabulary courses work: new
words in order of frequency, three grades for every answer, reviews on the
day an algorithm sets, and a drill of the words that went wrong. One engine
serves two ways in: the `vocabbot` program in a terminal, and a plugin for
Claude and Codex that runs the same session in a chat.

## How learning works

- **A day.** A day has three stages, as in SuperMemo: reviews due today
  come first, then new words, then the drill, where every card answered
  wrong comes back until the learner knows it. A drill answer leaves the due
  date unchanged.
- **The learner's choice.** The learner moves between the stages at any
  moment: to the new words while reviews wait, back to the reviews, to the
  drill. The day's number of new words is a start: more words join the day
  on request, as many times as the learner likes. After the due reviews the
  next round brings the cards with the nearest later dates, and the
  algorithm counts the shorter gap in their next date.
- **Cards.** A card shows the Polish meaning, the part of speech and an
  example sentence with a gap; the learner types the word. A new word
  comes at once as such a question, as in SuperMemo, and the result of the
  first answer shows it in full: gender, forms, IPA, a link to a recording
  and an example. A right first answer marks the word as known before the
  course and brings one more new word. The plan's `new_words=show` shows a
  new word before its question. The day's counter starts every message of a
  session in a chat.
- **Grades.** The engine compares the typed answer with the entry by the
  rules of the language: right gives "wiem", a small slip (a diacritic, one
  letter, the article, the case, another form of the word) gives "prawie",
  anything else "nie wiem". "Znałem" marks a word the learner knew before
  the course, at its first showing. The drill has two grades: "wiem" takes
  the card out and any other answer keeps it there.
- **Order of reviews.** The most overdue cards come first; the plan's
  `review_order` changes it to the order of the course, the shortest
  intervals first or a random order.
- **The card's controls.** As with SuperMemo's "Check", the learner may say
  "pokaż" instead of typing, see the answer and grade the recall; a typed
  word is still graded by the engine, and the plan's `answer_mode=type`
  leaves typing only. A sentence's translation comes when the learner asks
  for it, or at once with `sentence_translation=shown`. Every result says
  when the card comes back ("wróci za 4 dni"), unless `next_review=hidden`. A
  word the learner does not want goes aside and comes back on request, and
  the learner's own note for a word appears with its answer.
- **Schedule.** FSRS-6 sets the next review for a recall of 90% by default,
  the target of the SuperMemo courses. Its results match py-fsrs 6.3.3.
- **Several languages.** New words a day are shared between the languages by
  weight; the languages are interleaved in one session; one concept is not
  introduced in two languages within three days; an answer given in another
  language being learned is recognised and named.
- **Plan.** Languages, levels, weights, new words a day, days off, the target
  recall and the order of languages are settings of a plan. Every change of
  the plan or of the learner's own words is a preview first and is written
  only after the learner's yes.
- **Devices.** An account on a VocabBot server is the learner's record on
  every device. `vocabbot login` links a computer's learning folder to it,
  and the folder then synchronises by itself and works without a network;
  the server's session page (`/app/`) runs the same session in a phone's
  browser without a chat application. A session begun on one device goes on
  on another at the same card.

## Courses

| Language | A1 | A2 | B1 | B2 |
|---|---|---|---|---|
| English | 1079 | 1253 | 1919 | 2035 |
| German | 1016 | 1162 | 1724 | 1747 |
| Spanish | 1009 | 1126 | 1713 | 1733 |
| French | 1004 | 1140 | 1704 | 1731 |
| Italian | 1003 | 1134 | 1681 | 1758 |

The courses are built from open data: the levels from the CEFR-J
Vocabulary Profile, the translations, genders, forms and pronunciation from
Wiktionary (through kaikki.org), the order from wordfreq and the example
sentences with Polish translations from Tatoeba. Claude Sonnet 5.5 reviewed
the meanings and equivalents of all 6600 concepts, and the Italian
equivalents of 6700 concepts when Italian joined the courses. In a sample of 1200
entries, 1.2% had a wrong meaning, equivalent or gender. Where Tatoeba has no
sentence for an entry, Claude Sonnet 5.5 wrote one and the example review
checked it, so 99.5% of the entries have an example. 81.5% of the entries link
a recording on Wikimedia Commons (Wiktionary, Lingua Libre), from 98–99% of
the English entries to 27–79% of the Italian ones. Learners add their own
words from any text.

## Install

The plugin, in Claude Code:

```
/plugin install vocabbot --marketplace dithiothreitol/vocabbot
```

The plugin, in Codex CLI:

```
codex plugin marketplace add dithiothreitol/vocabbot
codex plugin add vocabbot@vocabbot
```

The terminal program, with uv:

```
uv tool install vocabbot
vocabbot init --lang de:a1-a2 --lang es:a1-a2 --new 10
vocabbot drill
```

Python 3.10 or later is the only requirement. [QUICKSTART.md](QUICKSTART.md)
walks through the first week.

The server for chat applications, with accounts for many learners, runs on
your own machine with Docker; Claude on the web and on the phone, ChatGPT,
Claude Code and Codex connect to it as to a remote MCP server. In Claude
and ChatGPT the session also shows a card with the day's counter and a button
that plays the word's recording. The same server has a session page for the
browser (`/app/`, also on a phone), an account page and the synchronisation
of learning folders (`vocabbot login`). [docs/server.md](docs/server.md)
describes the setup and the connection.

## Documents

- [QUICKSTART.md](QUICKSTART.md): installation, setup and the daily session.
- [plugin/README.md](plugin/README.md) and [plugin/PRIVACY.md](plugin/PRIVACY.md): the plugin and where the data stays.
- [docs/architecture.md](docs/architecture.md): the engine, the guarantees and the checks.
- [docs/development-plan.md](docs/development-plan.md): the staged plan and the results of each stage.
- [docs/pilot.md](docs/pilot.md): the protocol of the pilot before version 1.0.
- [docs/server.md](docs/server.md): the server for chat applications, its accounts and the connection.
- [CHANGELOG.md](CHANGELOG.md): the versions.

## Development

```
uv sync
uv run ruff format --check && uv run ruff check && uv run mypy && uv run pytest
uv run python scripts/build.py skills      # the engine and the rules in every skill
uv run python scripts/validate.py          # repository checks
uv run python scripts/session_test.py --check
```

`CLAUDE.md` describes the layout and the rules for changes;
`tools/courses/` holds the pipeline that builds the courses.

## Licence

The code is licensed under Apache-2.0 ([LICENSE](LICENSE)); the scheduler is
a port of py-fsrs (MIT). The courses are licensed under CC BY-SA 4.0.
[NOTICE](NOTICE) names the sources and their terms.
