1. Choose a model and language
The first load downloads the model to your browser — about 1 GB for the small ones up to 4–5 GB for the largest, shown next to each name in the list. It's cached after that, so future visits are instant. Bigger models are more accurate; smaller ones make more mistakes. A device with a stronger graphics card (GPU) can run the bigger ones — see what to expect.
Starting…
This browser doesn't appear to support WebGPU, so this page can't run an AI model locally. Try the latest Chrome or Edge on a desktop or laptop with a dedicated graphics card.
Session
Words your tutor has used, collected automatically as you chat. Saved only in this browser (nothing is sent anywhere). Tap "Explain" for a quick meaning.
Pick a real-life situation. The tutor plays the other person and stays in character — you have a short list of goals to complete.
Your turn — type or press 🎤 Talk.
Listen, then say the phrase. Your browser's speech recognition checks it word by word — green = heard clearly, amber = close, red = missed. It's a rough guide to how clearly you were understood, not a precise accent score.
This browser doesn't support voice input, so use the text box instead — everything else still works.
The AI conversation runs entirely on your device. Your vocabulary list is saved in this browser only (via local storage) so it's there next time — nothing is sent to a server. See our Privacy Policy for details.