Media
3 items
Description
Your voice runs the browser: speak commands on any site, dictate into any field, control pages with hand gestures. Don't type — just speak, anywhere on the web. B Agent turns your voice into clean, punctuated text in any text box on any website: Gmail, Notion, GitHub, ChatGPT, your CRM. Fillers removed, self-corrections followed ("meet at 5 — actually 6" becomes 6), and your names and technical vocabulary spelled right. YOUR AI TEAMMATE IN EVERY TAB Open the side panel to chat with your agent, see your tickets, and attach the page you're looking at — "explain this error" means the error on your screen. MEETING NOTES WITH NO BOT IN THE CALL One click on your Meet, Zoom or Teams tab records that tab's audio, builds a live transcript, answers questions mid-meeting, and files the minutes and action items into your project when you stop. HOW IT CONNECTS Requires a B Agent account and server (self-hosted or your company's instance). Pair in ten seconds with a one-time code — no password ever touches the extension, and you can revoke any device from Settings at any time. PRIVACY, PLAINLY Recording happens only when you press the mic — never in the background. Audio goes to your own B Agent server and nowhere else. No analytics, no trackers, no ads. The extension types only what you dictated; it never clicks, submits, or acts on pages for you. Good — that's just the last four empty fields on the Privacy practices tab. Paste these in (each is written to match what the code actually does, which is what reviewers check): **alarms** > Used solely for user-requested spoken reminders. When the user says "remind me in 20 minutes to check the oven," an alarm is scheduled and, when it fires, the reminder is spoken aloud via text-to-speech. Alarms are created only by explicit user voice command, stored locally, and deleted after firing. No background polling, tracking, or scheduled data collection. **scripting** > Used to inject the extension's own dictation/voice-command content script in two cases: (1) into tabs that were already open before the extension was installed or updated, so voice input works there without requiring the user to reload every tab manually; (2) to re-inject the current version after an update, replacing the stale copy. Only the extension's own bundled script (dist/content.js) is ever injected; no remote code, no third-party scripts. **tabs** > Used for voice-driven navigation and tab control: "go to youtube.com" (navigate the current tab), "switch to the Gmail tab" (find an open tab by its title and focus it), "new tab / close tab / mute tab," and zoom commands. Also used to track which tab is active so voice commands and optional hand-gesture input act on the page the user is actually looking at. Tab titles/URLs are read only in-memory for matching the user's spoken request; they are not stored or transmitted. **tts** > The assistant's voice. All spoken feedback uses Chrome's text-to-speech engine: confirming actions ("Opening youtube"), answering the user's spoken questions, reading page summaries the user requests, speaking reminders, and asking clarifying questions ("I can see Sign in and Basket — which one?"). Speech is generated locally by the browser's TTS engine in the user's chosen voice and language; nothing is recorded or transmitted.