Lingary

What it looks like

Nothing here is a generic lesson. Every mission, every word and the difficulty itself come out of what you personally said in the last conversation — which is what makes it a loop rather than a course, and what moves your level.

Your level

A1→A2→B1→B2→C1

Every lap round moves it up, and the next conversation is pitched there.

1 You talk to it

2 It finds the gaps

3 You add your own

4 You learn them

5 You clear the missions

Talking to it

• 🎙️

  Real-time voice

  A spoken conversation over OpenAI Realtime or Gemini Live, not a recording you submit. Cut in mid-sentence and it stops talking, the way a person would.

• ⌨️

  Or type instead

  The same tutor in text mode, switchable inside a session. Handy on a train, or when you'd rather not talk out loud.

• 🐱

  An avatar that reacts

  It lip-syncs to whatever is being said and shifts expression with the mood of the conversation — listening, thinking, pleased, unimpressed.

• 💬

  Subtitles, if you want them

  Live captions of what the tutor just said, toggled per session. Off by default so you have to listen first.

• 👆

  Tap any word for seven things

  Translate, hear it spoken, see examples, break down the sentence structure, get the grammar rule, ask a follow-up question, or save it to your deck.

• ⏱️

  It waits, then nudges

  Go quiet and a countdown appears rather than the tutor talking over you — extend it if you're still thinking. A small strip shows how many exchanges are left before the report has enough to work with.

• 🔖

  Every lookup is kept

  Everything you asked about mid-conversation lands in a list you can reopen later, answers and all — so a word you looked up last week isn't lost.

• 🔀

  Switch language mid-session

  Learning two things at once? Change which one you're practising from the toolbar without ending the conversation.

Six session types: free conversation, level assessment, word recall, grammar focus, mission practice and mission refresh.

How it decides what you practise

• 🧭

  Your level, from a conversation

  The first session is a chat, not a multiple-choice test. It settles on a CEFR level from A0 to C2 — with sub-levels — and adapts how it speaks to you from then on.

• 🎯

  Missions from your own mistakes

  Ten kinds — grammar, vocabulary, synonyms, pronunciation, conversation, cross-language parallels, verb matrix, story, inflection and word-family drills — generated from what you actually got wrong.

• 🔓

  Stages, and a Big Talk

  Four practice missions unlock a longer open conversation that revisits everything from that stage. Fail it and the stage re-locks until you've redone the retry missions.

• 🎲

  Two ways to progress

  Adaptive follows your weak spots. Structural works through a 30-verb grid across three tenses and six persons, filling in cells as you master them.

• 📖

  Reading, not just talking

  Story missions give you an AI-written passage at your level with the same tap-to-look-up actions as live chat, then a few comprehension questions.

• 🎚️

  Your own tutor prompt

  Not happy with how it teaches? Replace the system prompt with your own text and the tutor follows that instead.

• 🎭

  Pick the tutor's manner

  It changes the wording, the patience, how bluntly mistakes get called out — and even the tone of your practice reminders, which come in fifteen variants per style.

• 

  Friendly

  Encouraging, lets small slips go.

• 

  Strict

  All business — corrects everything, praises little.

• 

  Sarcastic

  Half-lidded, and enjoys your mistakes a little.

• 

  Drama queen

  Personally wounded by a wrong article.

• 

  Drill sergeant

  No sympathy, only repetitions.

Those are the avatar's own expressions, not stickers — it reacts in character while you talk. Switchable at any time in Settings → Learning; the livelier three ask for a confirmation first, because they are exactly as pointed as they sound.

The report after each session

• 📊

  Scored four ways

  An overall mark plus grammar, vocabulary, pronunciation and listening — with what you did well and the specific things to fix next.

• 🔍

  Words you didn't know

  Pulled out of your own sentences with translations and part of speech, and added straight to the deck. It also estimates your vocabulary size.

• 📈

  Trends over time

  Speaking pace and sentence complexity tracked across sessions, computed locally from stored metrics rather than asked of a model.

• 🎬

  Things to watch and read

  Two to four real YouTube channels, series, podcasts, books or songs per session, aimed at that session's weak spots. It never invents a URL — it hands you a search.

• 🎛️

  Switch off what you don't read

  Every optional block of the report has a toggle, and turning one off drops it from the prompt — so it isn't generated, and isn't paid for.

• ⚖️

  Grade with a different brain

  Chat offline but have the report graded in the cloud, or the reverse — and pin individual sections to either. Split runs fire concurrently, so you wait for the slower one, not both.

The report streams in: scores appear first, missions keep generating in the background, and a Continue button lets you leave without waiting.

Vocabulary

• 🗂️

  A deck that schedules itself

  FSRS spaced repetition with the four familiar grades. Cards arrive from conversations automatically, or you add them by hand.

• 📥

  Bring decks you already have

  Anki .apkg, CSV, TSV, Excel and JSON, with column mapping and a live preview. Quizlet exports come in through the TSV preset.

• 🍿

  Prep for something specific

  About to watch a film, read an article or play a game? Name it and you get the words and phrases you'll need, saved as their own group.

• ✅

  Answers graded by meaning

  Review checks whether your guess means the right thing, not whether it matches character for character.

• 🔔

  A word a day, and reminders

  Notifications from your own deck, worded in the voice of the tutor style you picked — fifteen variants each, so they don't get stale.

• 🕗

  A language per time of day

  Learning more than one? Set windows — German on weekday mornings, Italian in the evenings — and the app switches for you. Reminders can be turned on per language, not just globally.

• ✨

  Generate a set on a topic

  Ask for vocabulary on anything and it writes the cards: word, translation, an example sentence, the same sentence with a gap, and every inflected form that should count as correct.

• 🙅

  Words you already know

  Mark them once and they're excluded from generated lists for good, so you stop being taught things you learned years ago.

• 🧩

  Phrases and groups, not just words

  Multi-word expressions are their own kind of entry, and everything can be filed into groups — a deck per film, per trip, per topic.

Five ways to drill a deck

Vocabulary practice isn't only flashcards. The same words feed five different exercises, and three of them are written for you on the spot.

• 💬

  Practice with AI

  A conversation steered deliberately at the words you're behind on, so you have to actually use them.

• 🃏

  Flashcards

  Fill the word into a sentence rather than flipping a card, then grade yourself — Again, Hard, Good or Easy, with the next interval shown on each button.

• 📖

  Story

  A short passage written around your words at your level, with tap-to-look-up on every line and comprehension questions at the end. Pick how long it should be.

• 🔤

  Inflection

  Conjugation and declension tables with the cells blanked out. Fill them in and each one turns green or red — graded on the device, not by a model.

• 🌳

  Word Family

  One root, all its derivations, each with a gloss and examples you can save individually — then fill-in-the-blank sentences that use them.

Bring your own key — three of them, if you like

There's no subscription and no middleman. You point the app at the providers you want, with your own keys, and you pay them directly at their prices — the app takes no cut and runs no server of its own. And it isn't one choice but three: the brain, the voice and the ear are configured separately, so you can spend where it matters and save where it doesn't.

A settings screen shows which combination is actually live right now, since a pick that can't run — no key, no network, nothing downloaded — quietly falls back rather than failing. Keys are kept encrypted — in the iOS Keychain, or under an Android Keystore key — tied to that one device, and are only ever sent to the provider you chose.

Or no key at all

Offline mode replaces every cloud piece with one that runs on the phone, and there are two ways to get there. Either downloads once and then costs nothing, forever, with aeroplane mode on.

• Text conversation
• The session report
• Mission generation
• Spaced-repetition review
• Word bank & imports
• Reminders
• Every settings screen

What's weaker: voice becomes turn-based, so you can't cut the tutor off mid-sentence, and a small on-device model is coarser than a cloud one at grading and at drills. The CEFR verdict is deliberately distrusted offline — it records what it observed but won't move your level. Watch and read recommendations are cloud-only, because a 2B model cheerfully invents books that don't exist.

And the rest of it

• 📜

  Every session kept

  Full transcripts you can reread, with the same tap-to-look-up actions on any line you didn't understand at the time.

• 📐

  Goals in your own words

  Why you're learning comes up in conversation rather than a form, and what you said gets saved and fed back into later sessions. Editable whenever.

• 📏

  Grammar focus

  The rules you're working on, tracked as their own list and passed to the tutor so it keeps steering you into them.

• 📶

  Progress worth looking at

  Sessions, minutes spoken, words collected and words actually learned, plus what's due for review today.

• 🌍

  Several languages at once

  Each one keeps its own level, deck, missions and history. Add and remove them without disturbing the others.

• 🧹

  Deletion that means it

  Start one language over, remove it entirely, or erase everything — optionally including your API keys and the downloaded models. Each spelled out before you confirm.

• 📊

  Skills broken out

  The dashboard tracks pronunciation, grammar and listening comprehension separately, next to your goals, current missions and grammar focus.

• 🎯

  Retake the level test

  Think it pitched you wrong, or come back after months away? Reopen the assessment from settings and let it re-judge.

There's also a diagnostics export for when something misbehaves, and a live log viewer if you like watching the machinery.

Where your data lives

On your phone. There is no account to create, no backend holding your transcripts, and no analytics — your conversations, vocabulary and progress are a local database, and deleting the app deletes them.

In cloud mode your side of the conversation is sent to the provider you picked, because that's how it answers you — subject to their terms, not mine. In offline mode nothing leaves the device at all.

Where it's at

In open beta on TestFlight and on Google Play, waiting on App Store review. It's usable day to day — that's what it's being built against — but expect rough edges, and expect things to move.

On iPhone and iPad you'll need Apple's TestFlight app; the link walks you through it, and it runs on iOS 17 or newer. On Android it installs from Google Play like any other app, on Android 8.0 or newer. Offline mode wants a recent device — the models are large and it holds several in memory at once.

Found a bug? Missing something?

Write to me. It's one person building this, so there's no ticket queue and no support desk — email lands directly and gets read.

If it's a bug, these make it fixable rather than guessable:

• What happened and what you expected, and what you did just before.
• App version and build — Settings → About shows both.
• Your device and its iOS or Android version, and whether you were in cloud or offline mode.
• The diagnostics export if the problem is hard to describe — Settings → About → Export diagnostic data. It stays on your phone until you attach it, and you can read it first.

Nothing is sent from the app itself. There's no crash reporter and no analytics, so a problem I'm not told about is one I don't know exists.