Media
5 items
Description
Use your voice to click, scroll, search, dictate, edit and manage tabs in Chrome. Simple commands can act on partial speech. Talk to Chrome like a person sitting next to you. Hold one key, say "open the comments on the second story", let go. Yap looks for the link on the page and clicks it when it can identify the target. Put your cursor in a supported, non-sensitive text field and talk to dictate. "Change Thursday to Friday" and "delete last sentence" work as edits. What you can say • Page: "scroll to reviews", "click annual billing", "open the docs link in a new tab", "choose India", "play", "press enter" • Web: "go to example dot com", "search wikipedia for Alan Turing", "look up lofi beats on YouTube" • Dictation: cursor in a field, speak normally. "new paragraph", "change X to Y", "delete last word", "undo" • Chains: "search YouTube for TypeSafe Jev and play the first relevant result", "click Reply, type: Thursday works for me. Send it." • Tabs: "switch to the Vercel tab", "group all the Google tabs", "mute everything except Meet", "close duplicate tabs", "undo" Why it feels instant Yap doesn't ask a chatbot to write a plan. It shows a small model (Jev, from TypeSafe) selected page controls and asks typed questions — which element, is it a command, is it risky. Simple commands can run locally from partial speech. Speech recognition and target selection can still make mistakes; check important actions. Two ways to talk • Hold to talk: hold the right Option key (configurable), speak, release. • Ambient Mode: double-tap it and just keep talking while you work. Yap ignores conversation that isn't aimed at the browser and re-arms after every command. Say "stop listening" to end it. Built to be trusted • Password and card fields are never voice-filled or read. • Purchases, payments and reservations are blocked outright; send, delete, archive and sign-out ask you to say "confirm". • Closing more than three tabs asks first. Everything tab-related has "undo". • Semantic requests can include the spoken command, page title and address, short control labels and nearby text; they do not include full page HTML. In 0.4.8 and earlier, some existing non-sensitive field contents can also be sent. When speech-alternative correction is enabled, alternative dictated transcripts are sent to the chosen model provider. It is opt-in for new installs from 0.4.9 onward; existing preferences remain as set. See the privacy policy and Settings for controls. • Per-site blocklist for banking or work tools; ambient sessions time out on their own. • A decision log in Settings shows exactly what the model answered, how confident it was, and what it cost (about a tenth of a cent per hundred commands). Setup in one minute Enable the microphone and start talking. No account or key is needed for Yap's hosted model access, which has a fair-use limit of 60 requests a minute. You can use your own Vercel AI Gateway or TypeSafe key, subject to that provider's terms and limits. Works in Chrome 139+. English first; other speech languages selectable.