Solo
Speak into the phone in your own language; hear the translation in another. Useful at counters, taxis, anywhere you are alone with a question.
Translator
Speak across languages, from one browser tab.
A live translator with four shapes — solo, share-a-phone, multilingual room, paste-and-read. Bring your own OpenAI key and the whole stack runs from your browser.
Live
Why this exists
The translator apps on every phone are good at three-second exchanges and bad at everything else. They want accounts, send your audio to whichever cloud they prefer, and lose the plot the moment you try to host a meeting where four languages are in the room.
Translator splits the problem into the four shapes a real conversation actually takes. Sometimes you are alone, asking a question in a language that isn't yours. Sometimes you and one other person share a single phone, taking turns. Sometimes a group is in a room together, each speaking their own language. Sometimes you just need to read a long passage with the original alongside.
It is for travellers, multilingual teams, journalists, researchers, family members across diasporas — anyone who wants the convenience of a live translator without sending every word they say into someone else's logs.
How it works
You drop your OpenAI key into Settings, pick a mode, and start. Voice is captured in your tab, sent straight to OpenAI's transcription endpoint, translated, and synthesised back to speech — all between your browser and api.openai.com. Powerful Matter is not in the loop.
Room mode reuses the same peer-to-peer architecture as Secure Chat: each participant joins a passphrase-protected room, peers find each other directly over WebRTC, and each tab translates against its own key. Nobody has to share a key, and no audio passes through a central server.
Privacy in one paragraph
Your key is stored in your browser only. Audio, transcripts, and translations move directly between your tab and OpenAI; we don't see them. In Room mode, peer-to-peer connections carry the audio you choose to transmit; the signalling service that helps peers find each other only sees encrypted handshake material. We don't run a translation server, a logging service, or an account system — there is nothing of yours to leak.
Four modes
One shape per kind of conversation.
Duo
Share a single phone across a table. Tap to switch direction; each side speaks and hears their own language. Ideal for one-on-one meetings.
Room
Host a multilingual meeting from a passphrase. Each participant speaks their own language; everyone hears it in theirs. Peer-to-peer.
Reading
Paste a passage and get a clean side-by-side translation. Useful for articles, legal text, or anything where you want both languages visible.
Voices
Pick the TTS voice that suits the conversation. Multiple voices per language, configurable in Settings.
Auto-detect
Don't bother picking the source language — Whisper detects it. Just set the output and start speaking.
Things people ask before they trust this with a meeting.
Real conversations, real concerns. Honest answers below.
Voice is captured in your browser, sent directly to OpenAI for transcription, and the result returns to your tab. Powerful Matter does not see, store, or proxy any audio — open the network tab and verify that traffic goes to
api.openai.com.
Anything OpenAI supports for transcription, translation, and speech synthesis — that's around 50 languages today, including most major Latin-script, CJK, Arabic, Cyrillic, and Indic languages. Auto-detect handles the input side; you pick the output.
Translator itself is free. You pay OpenAI directly at their published rates for transcription, translation, and TTS — no markup, no subscription. A typical conversation costs cents.
Each participant joins a passphrase-protected room from their own browser, sets their language, and speaks. Each peer's translation runs against their own OpenAI key in their own tab — keys are never shared. Audio is exchanged peer-to-peer.
Yes — voice transcription, translation, and TTS all need to reach OpenAI. The page loads from us; everything else goes between your tab and OpenAI's API.
Yes — Reading mode is built for that. Paste a passage and you get a clean side-by-side translation. Useful for news articles, legal documents, anything where you want both languages visible at once.
Pick a mode. Speak. The rest is plumbing.
Drop in your OpenAI key, choose solo, duo, room, or reading — your tab does the rest. Nothing routes through us.