Conversations that fork
Fluency is handling the unexpected. Every turn has several paths, and your answer picks one.
- Your reply branches the story
- Same scene, different route
- The character adapts to what you said
Hold the mic and speak. Nothing moves forward until you do.
iOS · JLPT N5–N1 · Works offline

Five things the drill-and-memorize apps won't do.
Fluency is handling the unexpected. Every turn has several paths, and your answer picks one.
Hold to speak, release to be heard. The rhythm of a real conversation, not a tap-through quiz.
You don't need to know where you're weak. Every sentence you speak is analyzed in the background.
Hints fade, then disappear. At the top levels the AI writes new material just for you.
N5–N1 with furigana and bilingual examples. Build sentences aloud and the AI scores them live.

Order something complicated or keep it simple. The barista replies differently.

The AI coach pulls today's review set from your own error history.

A rush-hour coffee order, a ticket at Hakone station, a lost stranger asking directions.
Every sentence is diagnosed the way a teacher would, then turned into practice.
Word forms and conjugation
Structure and particles
Whether the wording is natural
Context and politeness
Pronunciation accuracy
These ship with the app. Type any other topic and the AI builds the scene, illustration included.
Order something complicated, under pressure
Fares, top-ups, fixing a mistake
Directions, landmarks, polite forms
Rental terms you can't afford to misread
Engineer interview, keigo under pressure
Izakaya small talk, a leak to report…
Type any topic and the AI writes the scene
Each step needs you to speak before it opens the next.
Pick your level the first time you open the app.
Word shadowing and grammar, all voice-driven.
Handle a branching conversation out loud.
SRS practice generated from your own mistakes.
| YuiTalk | Flashcard apps | Chat-only apps | |
|---|---|---|---|
| Voice is the core interaction | |||
| Branching, unscripted dialogue | |||
| AI-generated custom scenes | |||
| Five-layer error diagnosis | |||
| Reviews built from your errors | |||
| Offline speech recognition | — |
I had the N4 vocabulary and still froze. Speaking every day is what fixed it.
I'm moving to Japan for work. Branching scenes are the closest thing to real conversation I've found.
Years here and I still hesitate. The custom scenes match my actual life.
No. The neural recognition engine runs on your device and is tuned for Japanese, so it works on a bad connection or none at all.
N5 through N1. Every word ships with a bilingual example, audio and furigana.
A chatbot follows a script and accepts whatever you type. YuiTalk is voice-first and runs on a branching dialogue graph, so what you say out loud changes where the conversation goes.
It analyzes each sentence for morphology, syntax, lexis, pragmatics and phonology. Errors feed your weak-words notebook and the spaced-repetition deck built from your history.
YuiTalk is iOS for now. Scenes and features are still expanding.
Practice the one thing other apps skip: opening your mouth.