MirrorCaption VS ShellNewHandler

MirrorCaption's answer

MirrorCaption puts live captions and live translation on screen while people are still talking — not after the call, and not after the sentence ends. Words stream in as they are spoken, so both sides of a conversation read what is being said in their own language within about a second.

A few things that are unusual about how it does that:

• No bot, no plugin, no download. It works with whatever audio is there — Zoom, Teams, Google Meet, Webex, a phone on the table, or an in-person meeting in a conference room. Nothing gets invited into the meeting.
• Tap-to-compare highlighting. Touch a translated word to see the exact original word behind it, so you can check the translation instead of trusting it blindly.
• A vocabulary list you build as you go, which is why language learners and students keep it open beside their lectures.
• Automatic speaker detection, so a five-person meeting still tells you who said what.
• AI summaries while the meeting is still running, with action items and decisions pulled out at the end.
• Transcripts stay on your device, stored locally in your browser or app rather than collected into a cloud archive you have to trust.
• Interface in 25 languages, so the person you are helping never has to navigate an English-only screen.

MirrorCaption's answer

Most meeting AI tools are recorders: you invite a bot into the call, it sits there, and afterwards you get a transcript and a summary. MirrorCaption is built for the conversation while it is still happening.

• It works where a bot cannot go. An in-person meeting in a conference room, a phone on the table, a lecture hall, a clinic front desk. There is no call to invite anything into.
• It translates as people talk, streaming word by word instead of waiting for the sentence to end. Both sides read what is being said in their own language within about a second.
• Nothing to install. It runs in a browser, on your phone, or on your Mac or Windows desktop. No plugin, no download, no meeting bot.
• You can check the translation. Tap-to-compare highlighting shows the exact original word behind any translated word, so you are not asked to trust the machine blindly.
• Speaker detection keeps a five-person meeting readable — you still know who said what.
• Summaries run while the meeting is still going, with action items and decisions pulled out at the end, rather than arriving an hour later.
• Transcripts stay on your device, stored locally in your browser or app rather than collected into a cloud archive you have to trust.
• The interface itself is in 25 languages, so the person you are helping never has to navigate an English-only screen.

MirrorCaption's answer

People whose work goes multilingual mid-conversation and who cannot afford to slow down.

• Cross-border business teams — a support engineer in Manchester on a Teams call with a supplier in Shenzhen.
• International support and sales — a sales team running a demo for a prospect in São Paulo.
• Clinics and public services — a front desk registering someone who speaks another language.
• Universities and online courses — an international student following a fast-paced lecture in a second language, reading live translated captions in their own language while the lecturer keeps talking, then leaving with a full transcript and summary instead of half a page of notes.
• Language learners, who use the live vocabulary list and tap-to-compare highlighting to study from real speech rather than textbook sentences.

The common thread is that these are all situations where waiting for a translation, or inviting a bot into the room, is not an option.

MirrorCaption's answer

• Frontend: React 19 and TypeScript, built with Vite and styled with Tailwind CSS. The whole capture-and-caption experience runs in the browser (or Desktop Client).
• Realtime edge: Cloudflare sits between the browser and the speech engine, so the audio stream is handled at the edge.
• Backend: Supabase — Postgres plus edge functions.
• Mobile apps: Capacitor, with native audio capture written per platform so the app can keep listening in the background.
• Speech and language: a streaming speech-to-text engine that emits words as they are spoken, paired with LLM-based translation and summarisation.
• On-device storage: transcripts and session history live in the browser's IndexedDB on the user's own machine rather than in a server-side archive.
• Hosting: Vercel.

MirrorCaption's answer

I am a programmer at an international company, and MirrorCaption started with a problem I had in my own meetings every week. My foreign-language skills are good enough — I can follow a conversation. But following and catching every detail are not the same thing. In a fast meeting, with several people talking over each other, an accent I am not used to, a number or a name or a caveat mentioned once in passing, I would come away with the gist and a quiet worry about what I had missed. Asking someone to repeat themselves a third time is not really something you do on a call with eight people on it. Speech technology had finally become good enough to fix that, so I built the thing I wanted: live captions and live translation on screen while people are still talking, so I could read the detail I had only half-heard without interrupting anyone. I use it every day. That is the part that matters most for the product. Every rough edge I run into, I fix — including the small ones, the kind a team that only ever demos its own software would never notice. I am MirrorCaption's heaviest user, and I would not put something in front of you that I was not relying on myself the same afternoon.