Inspiration
I travel a lot, and I often need to look something up or work out a quick calculation when my hands are full or I can’t use my phone. Other days I spend hours in front of a screen, my eyes hurt, and I still have tasks to finish. I wanted to keep getting things done without looking at a screen. That’s how Nion started: a voice agent that can handle tasks for me, and the goal is for it to be able to set a reminder to book tickets next week, while also answering quick questions and fact-checks.
What it does
Say “Hey Nion” and talk to it like a person. It answers out loud using current information from web search, knows roughly where you are through your browser’s location, uses your name, and handles follow-up questions. You can interrupt it mid-answer, and your conversations are saved in a history. Answers are short and never include links, so you never have to look at the screen.
How we built it
I built Nion in Claude Code, following the hackathon’s plan-first skills: scope, PRD, and technical spec, then 9 small working build steps. The server runs on Node.js and Express and uses Claude Sonnet 5.5 through the Claude API with its web search tool. Chrome’s Web Speech API handles listening and speaking, and the browser’s Geolocation API plus OpenStreetMap turn your location into a city. There is no database, so history is stored in the browser. The project has 125 automated tests.
Challenges we ran into
The hardest part was positioning. Siri, ChatGPT Voice, and Alexa already offer voice, so Nion’s answer is to be fully screen-free, with concise replies and no links. On the technical side, Nion kept hearing its own voice through laptop speakers and mistaking it for me, and letting you interrupt without Nion cutting itself off took careful work. Chrome also misheard “Nion” at times, which once saved a wrong name. Spoken answers needed a different style from text, so I kept replies short and added a brief “one moment” for slow responses so it never seems stuck.
Accomplishments that we're proud of
I took this from idea to working proof of concept on my own. It took research, debugging, and a lot of back-and-forth testing, and positioning the product created many roadblocks along the way. Recording and publishing the YouTube demo was also a milestone.
What we learned
Building with AI means continually catching and correcting its mistakes. The skills package made the process interactive, and I played designer, product manager, and tester: walking through the scope, PRD, and technical spec and then building and testing. Constantly asking “Why are we doing this?” and “Why not that?” kept it from being just a coding session, and it gave me a strong foundation for building apps.
What's next for Nion.AI
The goal is a fully reliable, screen-free companion. With plugins for Gmail, Outlook, and the Microsoft Suite, Nion could draft and send emails or recall and organize your to-do list. I want it to travel with you and get things done while you never look at a screen.
Log in or sign up for Devpost to join the conversation.