Three Stanford-area engineers spent the better part of this year building what they hope will become the way people actually use voice assistants: a single button that summons an AI capable of seeing what you see and controlling the cascade of apps that run modern life.
Their company, Touchy, emerged from Y Combinator's latest cohort in August with an iPhone app that tries to solve a problem plaguing every voice assistant since Siri arrived in 2011. Users don't just want to ask questions anymore. They want digital help that can book the Uber, text the group chat, check the calendar, and remember that they prefer window seats without making them repeat themselves every time.
Whether Touchy can deliver on that promise in a market suddenly crowded with well-funded competitors remains an open question. But the app's approach reveals how a new generation of startups is rethinking voice interfaces now that large language models can actually understand context.
The software runs on iOS 18 and hooks into more than 1,000 different tools, according to the company. Users bind it to the iPhone's action button or camera control, then hold anywhere on screen to talk. The phone's camera becomes the assistant's eyes, reading signs, posters, or computer screens to understand what the user needs.
Touchy handles interactions most voice assistants still fumble. Lift the phone to your ear and it switches to private mode for sensitive conversations. Ask it to coordinate a dinner plan and it can check calendars across multiple people, suggest a restaurant based on past preferences, book a ride, and send confirmations to a group chat in one continuous exchange.
That continuity matters more than the individual features. The app maintains memory across requests, building what the founders describe as "a single, continuous conversation" rather than treating each command as isolated. It's the difference between having an assistant and having a very fast search engine that occasionally sets timers.
The three-person team brings unusually deep technical credentials. Rohit Agarwal is a computer science PhD student at Princeton PLI's Hazan Lab, with stints at Google DeepMind and Duolingo on his resume. Andrew Liu studied computer science, mathematics, and linguistics at UC Berkeley, where he worked on language documentation tools. Kevin Wang was a founding engineer at Eventual, a startup that recently closed a $30 million Series A, and contributed to the SkyPilot data engine during his Berkeley years.
Their timing puts them in direct competition with tech giants moving fast. OpenAI added video and screen-sharing to Advanced Voice for ChatGPT Plus subscribers in September. Apple gave Siri its own dedicated app in June. Google launched Gemini Live, its hands-free conversational assistant, in August. Each commands resources Touchy cannot match.
But the startup's founders argue that advantage cuts both ways. "Touchy is built to give you your attention back," they wrote in their launch announcement. "Ask it something, let it act, and get back to your day." The implicit critique: big tech companies want you inside their apps as long as possible. Touchy makes money by getting you out faster.
The company charges $14.99 for "Touchy Plus," an in-app purchase that unlocks full functionality. The App Store listing doesn't specify whether that's monthly or annual, though a promotional site mentioned one month free. For now, the basic app is free to download.
Early demand caught the team off guard. "We love all the hype around Touchy and are actually hitting rate limits with our providers!" Agarwal wrote on Product Hunt in early September. The comment suggested the infrastructure challenges that come with sudden traction, though the company hasn't disclosed user numbers.
The app landed at number 13 on Product Hunt the day it launched and shipped three updates in its first week, tackling multilingual support, augmented reality stability, and microphone quality. As of late September, the App Store showed three ratings averaging five stars, though that sample size makes any conclusion premature.
Touchy's website includes a manifesto with an audacious projection: by 2029, running an AI assistant smarter than current models will cost less than a music subscription. The claim cites research from MIT FutureTech and Epoch AI, organizations tracking the plummeting costs of inference as models become more efficient.
If that timeline proves accurate, the market for AI assistants could look radically different in five years. For now, Touchy faces the challenge every Y Combinator startup confronts after demo day: converting interest into sustainable growth before the funding runs dry or a larger competitor ships something similar.
The company hasn't announced funding beyond Y Combinator's standard backing. In the world of AI startups, that likely means a few months to prove the concept works at scale.
