Aalto: Talk to Your Browser

Aalto is a voice agent that lives in Chrome's side panel and acts on whatever page you are on. Press Alt+A and talk: it does the thing instead of describing it. It answers questions from the page you are reading, or from a live web search, and highlights where the answer came from. It dictates into any text box, and when you say "summarize this" or "make it more formal" it rewrites your text in place. It fills several form fields from one sentence, scrolls, jumps to sections and finds words on any page - including apps like Gmail that scroll an inner panel - and opens, switches and closes tabs by whatever you call them. AssemblyAI's Voice Agent API is the whole voice loop: speech recognition, turn detection, barge-in, routing across 18 browser tools, and speech back. There is no orchestration server; a small Cloudflare Worker only mints single-use tokens under a daily budget. A get_context tool lets the agent see what you see - the active tab, the box you are typing in, your selection and every tab by id - so "this", "it" and "the tab you just opened" resolve correctly. Conversations carry across pauses without starting over, and sessions pause themselves when you go quiet, because the connection is billed by the second. Safety is enforced in code, not in the prompt: Aalto will not submit a form it has not read back to you, and will not guess an option you did not say. It is built for people who fill in forms and move between tabs all day - support, operations and sales teams - and for anyone for whom typing is slow or painful. It works on any website with no integration. Try it without installing: the website runs the same tool definitions against a small browser inside the page, with a live panel showing every call the agent makes.