Speaking Conversation
A conversation, not a script.
You talk, an AI tutor talks back, out loud, in real time — a live voice model, not recordings. It picks up where you hesitate, offers a better way to say what you were reaching for, and keeps going.
It corrects while you talk
Each fix is recorded the moment it happens, not saved for the end — so even a call that drops leaves you the “you said → a better way” pairs you had already earned.
Captions are on by default
They are a comprehension aid, not a debug view. The tutor can also push a word or phrase onto your screen mid-sentence when hearing it isn't enough.
Metered by the server, not the browser
Every call has a ceiling — five minutes on Free, ten on Pro, fifteen on Pro+ — and the tutor is asked to start wrapping up two minutes before it. The count comes from the server's own clock rather than anything your browser reports.
A speaking lesson: a glowing orb pulses while the tutor talks, with the last few lines of the conversation captioned underneath.