Skip to content

Aria

Aria is a voice assistant that browses the web on your behalf. You speak a goal; Aria navigates real sites, reads back what matters, and acts — designed first for people for whom a mouse-and-screen workflow is a barrier.

The web assumes you can see it and point at it. Voice interfaces usually stop at search and weather. Aria closes the gap: a spoken request becomes real navigation and action in a live browser.

  • Voice in — speech is transcribed into an intent.
  • Plan — the agent decides the steps needed on the target site.
  • Browser use — it drives a real browser: navigate, read the DOM, click, fill, extract.
  • Voice out — results are summarised and spoken back, kept short and skimmable by ear.
speech ─▶ intent ─▶ plan ─▶ [ browser: navigate · read · act ] ─▶ spoken summary
  • An accessibility-first framing that shapes every design choice, not a bolt-on.
  • Combining voice and browser/computer use in one loop — two hard things together.
  • Summarising for the ear, not the eye: different constraints than a chat UI.

TypeScript · speech-to-text / text-to-speech · browser automation · LLM planning

A walkthrough video goes here — this demo is much more convincing watched than read.