B Agent

Full description (paste into "Detailed description") Don't type. Just speak. B Agent runs your browser by voice. One click on the microphone — or one hotkey — and you can dictate into any text box on any site, command pages out loud, chat with an AI teammate that sees the page you're on, and capture meeting notes without inviting a bot to the call. 🎤 Dictate anywhere Speak into email, docs, notes, team chat, project trackers, code hosts, and any text box on the web. Words stream in as you talk, with spoken punctuation ("comma", "new paragraph"), automatic capitals, your own vocabulary corrections ("use effect" → useEffect), a personal dictionary, reusable snippets, and number/case/accent cleanup rules you control. On stop, the take is optionally polished by AI into the right shape for the site — a casual message for chat, a proper email for mail, prose for long-form writing — and it learns your style from your own edits. 🖱️ Command the page hands-free Open sites, run searches, click buttons, fill fields, scroll, move between tabs by name, open and close tabs, zoom, send composed messages, and press Enter — all by voice. Ask questions and hear spoken answers, get page summaries, hold a follow-up conversation, set spoken reminders, hand over multi-step goals, file project tickets, and control meeting capture. Interrupt anytime — it stops talking and listens, just like a person. Commands also work in Hindi and Hinglish. 🤖 AI teammate in the side panel Chat with an agent that can read the page you're looking at (with your per-message consent), pick a project, manage tickets, and answer by voice in a natural, humanized voice you can restyle — rate, pitch, persona, 25+ languages, even a neural server voice. Talk to it entirely hands-free: speak, get an answer, speak again. 📝 Meeting notes with no bot in the call Say "take notes" on any video-call tab. B Agent captures the tab's audio plus your mic, attributes speakers from live captions, and on "write the minutes" produces a minutes document with action items filed as tickets. Works on-device without an account; paired, it also transcribes the other participants. 👋 Hands, claps, and quiet control Point to move the cursor and pinch to click with hand-gesture control (the camera runs on-device; closing the panel doesn't stop it). Double-clap to summon the mic. Hold Alt to talk briefly while muted. The toolbar glyph itself shows state — struck-through grey when idle, open green when listening — with the gesture hints in its tooltip. 🔒 Private by architecture Password, card, CVV, OTP, and account-number fields are never dictated into or read. A site deny-list silences B Agent completely on banking and any host you add. Your audio goes to your own B Agent server (self-hostable) — no third parties, no trackers — with transcript retention you control and a one-click history wipe. Private mode transcribes on your server instead of the browser engine. ⚙️ Yours to tune One-time mic/camera grant on install covers every site. Pairing is a short code typed once in the web app. Then: voices, sensitivity, clap and gesture tuning, AI persona and answer length, writing styles, dictionary, snippets, meeting defaults, themes, and keyboard shortcuts (the toggle shortcut dictates; double-click the icon opens the panel). Free to use for browsing, dictation, and gestures. Pairing a B Agent account adds the AI: polish, spoken answers, meeting minutes, and rewrites.