Language learning apps without an AI tutor

The phrase covers at least four different wishes, and they are not equally easy to satisfy:

• No AI tutor. You do not want a chatbot pretending to teach you, correcting you, or inventing grammar explanations.
• No generated content. You want the words and sentences you practise to come from your lesson or your reading, not from a model writing filler.
• No AI at review time. You want the same question to have the same answer tomorrow, not a different explanation on every visit.
• No notes leaving your device. You do not want your class notes, tutor conversations or personal writing sent to a third-party provider at all.

A vocabulary app can meet the first three and still send text to a model once. It cannot meet the fourth while also turning free-form notes into structured vocabulary, unless the extraction runs on your own device. That is a different product with different trade-offs, and it is worth knowing which one you are looking for before you compare apps.

One step, at import. When you paste or scan lesson notes and press the import button, the text goes to Google Gemini, which returns a proposed word list with translations, a short story using those words, and quiz sentences. You approve, edit or discard that proposal before anything is saved. The import is rate-limited and the request has a deadline, so a slow model response fails the import rather than hanging the page.

After you save, the model is out of the loop. Review questions come from your saved records and the quiz sentences created at import. The spaced-repetition intervals are calculated by the app. The story you can replay from your stories page was generated at import and does not change afterwards. If you open the same word next month, you get the same stored sentence, not a freshly invented one.

This is disclosed in the privacy policy, including that on-device photo and PDF recognition happens on your device and only the recognised text is sent for parsing.

You bring a lesson: a page of notes, a list from a tutor, a paragraph you underlined. The import proposes the words worth keeping. You choose. Each saved word arrives with the context you met it in, so you are not reviewing a bare translation pair.

Recaps then bring words back on a schedule. A word you know well returns later; a word you keep missing returns sooner. Some questions ask for a translation, some show a sentence with a gap, some ask you to pick the right form. The point is recall, not recognition, and the point of the schedule is that you stop reviewing what you already know every single day.

You can see the workflow without an account in the read-only Italian demo. It uses a fixed sample lesson, so you can judge the review experience before importing anything of your own.

A general chatbot is good at producing an explanation on demand. That is also the problem: the explanation is different every time, it can be confidently wrong, and it has no memory of what you actually studied last week. For vocabulary, the useful question is narrower: do you remember this word from your own lesson? A stored question with a stored answer is a better fit for that than a conversation.

There is also a cost argument. A tutor-style chat needs a model call for every exchange. A vocabulary app needs one call per import and none per review, which is why the import step can be a one-off cost rather than something that runs on every review.

LanguageRecap is not a course and does not decide what you should learn next. It will not teach you a grammar point from scratch, hold a conversation, or grade your pronunciation. If you want a guided path from beginner to intermediate, a course app or a teacher is the right tool. This app assumes you already meet language somewhere and helps you keep it.

It is also not fully offline. Import needs a connection; review of saved material is a normal web app and does not call a model, but it is not a downloadable offline pack.

Four questions separate a precise claim from marketing:

1. When does the model run? Once per import, once per lesson, or on every question? If the answer is not on the site, assume the worst case.
2. What is sent? Your notes, your review answers, your voice, or nothing? The privacy policy should name the provider and the data.
3. Is the output stored or regenerated? Stored output is stable and auditable. Regenerated output changes and cannot be checked once.
4. What happens if the model is wrong? You need to see and edit the proposal before it becomes study material.

An app that answers all four plainly is easier to trust than one that only says “AI-powered” or, in the other direction, “no AI”.

Does LanguageRecap use AI?

Yes, once per import, to turn notes into a proposed word list, a short story and quiz sentences. The model is Google Gemini. There is no AI tutor and no chatbot.

Does AI run while I review?

No. Review questions, quiz sentences and spaced-repetition intervals come from your saved records. The same word shows the same question later.

Can I use it without importing notes?

The demo is a fixed Italian workspace and does not import anything. To study your own material you import a lesson, which is the step that uses the model.

Is my imported text used to train a model?

The privacy policy describes what is sent to the model provider and for what purpose. Read it before importing sensitive material; ordinary lesson vocabulary is the intended input.

Which languages are supported?

Vocabulary review works for English, Italian, Spanish, French, German, Latvian and Swedish. Conjugation review is available for English and Italian. The demo uses Italian.

If the scoped version of “without AI” is what you wanted, create an account and import one lesson. If you need a tool that never sends notes anywhere, this is not it, and the honest answer is to keep looking. Our guide to choosing a language learning app, the Anki alternative comparison and the Duolingo comparison cover the rest of the decision.