The small idea of starting Noteshell

It all started around October 2025, when I was locking in to learn about AI systems. My routine was to read books, like Deep Learning by Ian Goodfellow (you can find more of what I’ve read on my shelf). Whenever a topic or term was hard to understand, I asked Gemini’s Deep Research about it, and it gave me a detailed, insightful report.

To understand those topics more deeply and remember them for longer, I took notes in Notion. Typing things out helped me remember. The reports also didn’t carry across chats, so I often lost track of them in old threads. And I couldn’t just copy Gemini’s text into Notion, because the two apps handle formatting and text differently.

That gave me the idea for a note-taking app where AI writes the notes, bringing that scattered workflow into one place. The notes could have visual, interactive parts that make them feel alive: a note about neural networks could let you adjust the weights, or show propagation in action.

But over time, I noticed something wasn’t clicking. The first sign was that I still didn’t feel it would help me. I also had doubts about whether it was technically feasible, and I didn’t think people would pay for another “AI wrapper”, an app that just calls APIs.

And all the way through, I kept thinking about how we could embrace the flexibility of AI. AI will become much, much more capable than it is today, so how could the app stay flexible enough for AI to handle more and more? That question kept running through my head. When we pitched it to a few people and got a lot of positive feedback, I knew I had to find another way forward.

The version closest to today’s Noteshell was the one from about five months ago. It came out of my friends and me brainstorming a different approach. That version leaned toward turning every file, link, and piece of information you had into reports and dashboards.

The problem with that direction was that we couldn’t attract many active early users, and the idea had drifted away from our original goal: to serve everyday people. So I took a step back and rebuilt from that version.

The first milestone of today’s Noteshell came when it could build reports, slides, videos, and dashboards natively, at a level I thought could truly make a difference for people. The final sign that this could work came when one of my friends used Noteshell to automate her interview preparation, and went on to get offers from Amazon and Microsoft. That was one of my biggest inspirations to keep pushing forward.

For more than five months, I’ve often found it hard to explain what Noteshell does, and how it’s different from other tools, without showing it. So I kept telling my friends to try it, betting they would love it. (I won that bet many times.)

In one sentence: Noteshell is a knowledge base with AI agents that build things like notes, reports, videos, dashboards, or even apps, and manage them on their own to help you reach your goals.

The idea is that people have goals and tasks, and they need tools to get there. Those tools are usually spread across many places, and some need to be shaped around the person using them.

Take my situation last year. Noteshell could give me a daily lesson on one AI topic as a note, with a flashcard after each one, and a dashboard to track all my learning progress.

Or take a bigger goal, like finding an internship. Agents could look up jobs and draft a plan for applying to each one, while another agent oversees it all and gives you a weekly plan to keep track of everything. There could also be a folder where agents prepare LeetCode questions and explanations, so everything is ready for you.

That’s what we want Noteshell to be: the best place to actually reach your goals, and, in turn, a place that makes learning and exploring easier and freer.

Noteshell is still early, and still finding its fit. But seeing it bring real value and results to our early users is a huge motivation. It keeps us believing in our core idea, pushing forward, and steering where Noteshell goes next.