Local and private by default
The best place for your AI to run is your own computer, not someone else’s cloud. Your work stays in the room.
Manav Sehgal
Founder, Orionfold · builder for twenty-five years
Hi, I’m Manav. For twenty-five years I helped the biggest companies in the world put powerful technology in front of millions of people. Now I do the opposite of big. I put that same frontier in the hands of one person at a time, starting with you.
That is the whole mission. The frontier of AI should not live only in giant clouds rented by giant companies. It should fit on your desk, run on your terms, and answer to you.
Why I do this
For most of my life, the rule was simple. The best technology was big, expensive, and lived far away in someone else’s data center. You could use it, but you could not own it. I spent years on the inside of that world, and I kept noticing the same gap. The tech got more amazing every year, and yet it reached real people slowly, if at all.
Then something changed. Open AI models caught up to the locked, paid ones. A computer small enough to sit on a desk got powerful enough to run them. For the first time, one person with the right tools could do what used to take a whole team and a cloud bill. I could not unsee it. So I left to build for that person.
The path here
My path was not a straight line. It was the same idea told in very different rooms. It started at Xerox PARC in the late 1990s, where I helped build one of India’s first systems for storing and sharing documents across distance.
At AWS I helped bring the cloud to businesses across India, and helped stand up the tech behind the country’s pandemic response. At Amazon I led product for Alexa’s conversational AI, then worked on the foundation models that ship today as Amazon Nova, then helped bring generative AI to the store millions of people shop in every day. After that I led the team helping frontier AI labs like Anthropic and chipmakers like NVIDIA build on AWS.
Today that NVIDIA thread runs the other way. My reference machine is the NVIDIA DGX Spark, a small AI supercomputer on one desk, and the book, models, and tools are all proven on it. You can see that work gathered on the DGX Spark page.
Every step taught me the same lesson, a little louder each time. The most powerful technology in the world is worth almost nothing until it reaches the person who needs it. In June 2026, after nearly nine years at Amazon, I left to do the one thing all of it pointed at.
Then I spent sixty days building this startup the AI-native way. One person, one desk, research and software and books and operations, all with AI helpers alongside me. What I learned surprised me. AI did not make me smarter. It let me use different parts of my thinking together, and hold the thread between them. I wrote that down in Limitless, Without the Pill. The short version is a sentence I keep coming back to: I do not write every note anymore, I conduct.
That sentence is now a product. Orionfold Flow is the Mac app I am building so anyone can work that way, in the place the work actually happens: the document. Everything before it taught me what to build. Flow is me building it.
Things I’ve helped bring into the world
Not a resume. A timeline of the real, public things I had a hand in shipping, and the one I am building now.
Late 1990s
Xerox PARC
Helped build one of India’s first systems for storing and sharing documents across distance.
2017
AWS, India
Brought cloud computing to businesses across India, and helped stand up the tech behind the country’s pandemic response that reached 100 million people in 40 days.
2021
Alexa Conversations
Led product for Alexa’s conversational AI and grew it from 1 language to 9.
2024
Amazon Nova
Worked inside Amazon AGI on the foundation models that ship today as Amazon Nova.
2024
Generative AI for the store
Helped bring generative AI to the Amazon shopping experience that millions use every day.
2025
AWS Frontier AI
Led the solutions work helping frontier AI labs build on AWS, with partners like Anthropic and NVIDIA.
2026
Orionfold
Open AI software, custom models, and books you can run, all from one small desk.
2026
Orionfold Flow
Building the Mac app that lets one person conduct their documents with AI, the way I learned to work.
What I believe
Local and private by default
The best place for your AI to run is your own computer, not someone else’s cloud. Your work stays in the room.
Skill beats staff
One person with AI helpers can now do what used to take a whole team. I build the tools that make that real.
Open, not locked in
Open AI has caught up to the paid kind. I ship open software, open models, and books you can read and run.
The frontier is finally small enough to hold. Let’s put it in your hands.
I write a short letter every month on where AI is going and why a desk you own beats a meter you rent, and the proof that a desk can do real AI is in the receipts. The full story of why I left, and what the name Orionfold means, is in Why I folded Orionfold, and what I am building next is in Limitless, Without the Pill.
The tool I am building now
Orionfold Flow is that idea made into a Mac app. A place to conduct documents with AI agency built in, where you stay the one who approves what changes. It is in development as a freemium subscription, and the waitlist is open.