Why I'm Building Mikipage

Author: Daniel Zhou · Date: August 6, 2026

The world does not need another note-taking app.

I know this better than most, because I've spent 25 years using them. Not trying one for a weekend and writing a review — actually living in them, a decade at a stretch, with thousands of notes I still care about. There are 50+ good note apps out there. Several of them are excellent.

So I didn't build one.

Here's what took me 25 years to notice: notes were never what I wanted. Notes are just what I'm able to write down in the moment. What I actually want out of them is two things — the distilled version, a handful of pages that say what I've come to think about a subject, and the one small buried detail whenever I need it. Every app I've used was good at storing the raw material and did nothing with it. The distilling stayed my job, and I never did it. Nobody does.

Mikipage is my attempt at the missing half. You write notes; AI writes the pages — wiki articles that stay alive, rewritten as the notes underneath them accumulate. And since a wiki was never meant to be read by one person, the same thing works when your family or your friends are writing into it too.

That's the idea. It's also why I called it Mikipage and not Mikinote.

Here's the 25 years that got me there.

25 years of taking notes

2000 — MyInfo DB. I was still in China, I had reading notes piling up, and there was nothing good to put them in. So I built my own and called it MyInfo DB: a Microsoft Access database with fields for author, keywords, source, and a big text blob. It held my reading notes and, because I was that kind of person, every movie I watched and my thoughts on it. It was ugly and it ran on exactly one PC, but it did the one thing I actually wanted — let me find what I'd written six months earlier, with full text search on multiple metadata fields. I used it for years.

Then I moved to the US, and MyInfo DB stayed behind. Not deliberately — I just never migrated it. It's still sitting in an .mdb file somewhere, technically mine, practically gone. Remember that; I'll come back to it.

2010 — Evernote. Evernote solved the thing MyInfo DB never could: it was everywhere. Phone, laptop, browser. I moved in and stayed for over a decade, which is a long relationship by software standards. (The movie log went to Douban and never came back — my writing was already scattering across services, and I didn't think much of it at the time.)

What eventually ended it wasn't one thing. Pricing kept climbing past what it delivered, long-standing asks like Markdown went unanswered for years, and the app got heavier every release — more features I never opened, slower at the one thing I did.

But the specific thing that pushed me out was Evernote capping how many computers I could use. That's the moment the deal changed. The entire reason I'd left my own database was to have my notes everywhere; now the tool was metering exactly that, and charging me to lift a limit that hadn't existed the day before.

2023 — Obsidian, synced over Dropbox. Obsidian was a genuine relief. Plain Markdown files on my own disk. Fast. Local. Backlinks. A serious community. I put the vaults in Dropbox so my notes would stay in sync across machines, and I recommend the setup to people constantly — I still think Obsidian is one of the best-designed pieces of software in this category.

I'm still leaving.

What broke for me in Obsidian

None of what follows is a bug. They are all consequences of Obsidian being exactly what it says it is — a local-first tool for one person who enjoys organizing things. I am no longer that person, and I stopped pretending I was.

It lives on devices, and I have too many devices. Personal laptop. Work laptop. The family desktop in the living room, which is not mine to install software on. Phone. Tablet. To have my notes with me, Obsidian has to be installed and updated on every one of them — and on a few I don't control at all. There's no web access. If I'm on a borrowed computer, or a school computer, or someone's iPad, my notes simply aren't there.

Dropbox papers over the desktop half of this and falls apart on the rest. Mobile is where it really breaks: Obsidian on a phone can't open a vault living in Dropbox, so I'm left reading and editing my notes through a file browser, which is miserable. And it raises an awkward question — if I'm editing my notes in Dropbox anyway, what exactly is Obsidian doing for me?

It isn't only the app that has to travel. The plugin ecosystem is Obsidian's superpower and its tax: semantic search means picking a plugin, wiring up a model, and tuning it — on every device, again whenever it breaks. Power that doesn't travel with me isn't power I actually have.

Sync costs more than I think notes are worth. The proper answer is Obsidian Sync. One vault for $48/year is defensible. But I don't have one vault — and multiple vaults means the higher tier at roughly $100/year, capped at 10GB. That is $100 a year to have my own text files show up on my own phone. I can afford it. I just don't believe it, and I'd rather put that money toward something that deserves it.

Vaults made me file things before I could write them. Vaults are physically separate note collections, and they force a decision at the worst possible moment: before you've written a word. My setup was a "Personal" vault (reflections, book notes, parenting, quotes) and a "KitchenSink" vault (receipts, tax backups, dead projects). Then I scanned my driver's license. Personal document, or archived paperwork? I picked one vault. I no longer remember which, and I've never found it again.

Why not merge them? Because KitchenSink has thousands of notes I never want to see. Merged, every search for Mini Cooper maintenance tips returns Mini Cooper service receipts. So I keep the split, and I keep paying for it — a small tax on every single note, forever, in exchange for search results that are slightly less bad.

Sharing is an afterthought. Roughly a third to half of what I write, I eventually want someone else to see — my wife, my kids, friends, coworkers. Obsidian's answer is a shared vault, which means other people can edit my notes, or Publish, which pushes a copy to a website. Both put me in the copy-maintenance business: an original in my vault, a duplicate in the shared vault, and me manually keeping them in step. That's not sharing. That's bookkeeping.

And it keeps getting bigger. Especially on mobile, where the interface carries the full weight of a desktop power tool. The 20% of advanced capability kept getting in the way of the 80% I actually do.

What else I looked at

Notion — Clearly excellent, and clearly not for me. Databases, kanban boards, wikis, projects, tasks. It's an operating system for a company, and I don't want to run a company in my notes. Every extra surface is another decision to make before I can write a sentence.
Google Docs — Not a notes tool, whatever we use it for. Try imagining 400 Google Docs. Now find the one about your kid's swim schedule from last spring. Search over a Drive folder is not the same as retrieval over a body of writing.
Apple Notes — Genuinely good, and genuinely welded to Apple. My family isn't all-Apple, my work machine isn't Apple, and I don't want 25 years of writing to depend on staying inside one company's hardware.

Notes were never what I wanted

Every tool above is a filing cabinet. MyInfo DB, Evernote, Obsidian, Notion, Apple Notes — all of them store what I wrote and help me get it back. They differ on price, polish, and how the drawers are arranged. Not one of them has ever written anything. After 25 years and thousands of notes, turning my writing into something I could actually use was still, in its entirety, my job.

And it's a job that never gets done. Nobody re-reads four years of journal entries to see what changed. Nobody assembles the 60 scattered notes about a kid's schooling into one clear picture. I have never once sat down and worked my reading notes into what I actually think about a subject — and that was the entire reason I started keeping them, in 2000, in Access. The raw material was all there the whole time, beautifully filed, unread.

What I wanted wasn't a better cabinet. It was a wiki about my own life that keeps itself up to date.

Andrej Karpathy's "LLM Wiki" idea named this the moment I read it: stop organizing notes by hand and let a model maintain a living wiki over them. It's the right idea, and I'm not the first person building on it. But every implementation I found was single-player, needed real setup on every machine, assumed a monthly AI subscription, and — the part that worried me most — let AI rewrite the notes themselves.

A wiki is not a personal artifact. Wikipedia works because many people contribute and someone weaves it together. But traditional wikis put that weaving on the contributors: learn the syntax, respect the page structure, don't step on each other's edits, argue in the talk page. That barrier is why the shared family wiki, the shared team wiki, the shared club wiki almost always dies after three weeks.

Take that barrier away and something changes. Everyone just writes notes, in their own words, in their own space — and AI does the weaving. Nobody edits anybody else's writing. Nobody has to agree on structure. The page updates itself as new notes arrive from any of you.

That's a "miki": a wiki, maintained by AI, fed by however many people are writing into it. The note-taking underneath still has to be genuinely good — I'm not going to write in something that annoys me — but the notes are the raw material, not the reason.

What I want Mikipage to be

Not a feature list — the things I won't compromise on, in the order the rest hangs off them.

1. It's a wiki, not a notebook. The pages are what I'm after; the notes are what they're made of. Writing has to feel good or nothing else ever happens — but a lovely editor with nothing built on top of it is just a nicer filing cabinet, and I've owned several.

2. I write. AI writes back — and the line between us is absolute. Getting the thought out of my head is my job and no one can do it for me. Everything after that is what machines are good at. So there's a hard line down the middle:

What I write is mine and stays untouched. No one else edits my notes. AI doesn't edit my notes. They sit there exactly as I wrote them, forever.
What AI writes is alive. Pages are AI's territory — generated, updated, rewritten as the notes underneath them change.

Your notes are sacred; the wiki is alive. Every AI writing tool I've tried creates a low-grade anxiety about what it just changed. This one is structurally incapable of it, and it's the piece I care most about.

3. Everyone writes. AI synthesizes. A third to a half of what I write is eventually for someone else. I want to write it once and point it at the right people — my wife, the family, a group of friends, the whole internet — with no copies and nothing to keep in sync. Sharing a note doesn't hand anyone a copy or a pen; it feeds the group's pages. This is the part I couldn't find anywhere, and it's most of why I'm bothering.

4. I can find it again — and find what I forgot. Two different jobs. Retrieval: I know I wrote it, get it back, even when I don't remember the words. Discovery: show me the pattern across four years of writing that I never noticed. The second is the one that makes any of this worth it.

5. I can write from anywhere. When I want to write, I write. Any browser, any device, borrowed or mine, nothing to install. The browser is the app. Local tools can talk to it, but they're clients — the cloud is where the writing lives.

6. Simple. Apple's one-button mouse. Google's one search box. A small number of things, done well, out of the way. Every feature that isn't essential makes the essential ones harder to find. When something powerful but niche needs a home, it goes to the API — not the screen I look at every day.

7. Frugal AI. A living wiki means re-synthesis, and re-synthesis costs real money. So don't reprocess unchanged notes. Cache. Batch. Being cheap to run isn't housekeeping — it's the engineering that makes the next point possible at all.

8. Affordable. The lesson of 25 years is that note apps get expensive slowly. So: every paid tier is feature-complete. Paying more buys more usage, never unlocks a button. Priced at what a family can keep paying indefinitely without thinking about it.

The last two come last on the list and first in importance. They aren't features I chose. They're what I owe anyone who puts their writing here — including me.

9. My notes are mine. Full export, Markdown throughout, a complete open API to access the app programmatically. If Mikipage stops being worth using, you should be able to leave with everything, easily.

This is the MyInfo DB lesson. Those notes aren't locked away by a hostile company — I own the file. But it's a proprietary format from a dead workflow, and the friction of getting the writing back out was, for 25 years running, always slightly higher than the value of doing it today. That's how you lose a decade of your own thinking: not to a villain, to friction. Getting your notes out has to be easy on the easiest day, or it never happens.

10. Security and privacy is job #1. Non-negotiable, no asterisks. Encrypted in transit and at rest, on AWS. My own family's private writing is in here, and my kids' writing will be. That's the standard, and it doesn't move.

I should be honest that most of that list is unremarkable. Simple, affordable, portable, private, works everywhere — plenty of good tools already manage all five. None of it is a reason for this to exist; it's just the character I want it to have, and what I check against when deciding what not to build.

The narrow part is only this: AI writing living pages, out of notes written by more than one person, at a price a family can keep forever. I looked for a while, and I don't think anyone is doing all three. The tools where AI organizes for you are built for one person — one of them published a manifesto on why collaboration is a mistake. The tools built for sharing with your group still leave the organizing to you. And almost none of them are priced for a household.

Which is a decent argument that it's a bad idea, and I've considered that. I'm building it anyway, because I want to use it.

I'm building this for myself

Mikipage is what I want to spend the next 25 years writing in. My family will use it. My writing — the Evernote exports, the Obsidian vaults, and yes, eventually MyInfo DB — is going in. Which means the bar is complete satisfaction for one very demanding user, and I get to iterate fast precisely because that user is me.

It's worth saying plainly why now. This isn't something a small team could have built five years ago; it's a cloud service with an AI pipeline, a sharing model, and three separate surfaces (Web, API and MCP). AI-assisted coding is what brought that within reach of a team our size — which is a funny thing to have to admit about a thing whose whole premise is "let AI do the part machines are good at." Same argument, applied to itself.

And we are a small team, deliberately. I think it's the better deal: the person who reads your support email is the person who can fix the bug, and nobody here has to justify a growth curve before fixing something that's merely broken. What a small team owes you in return is a door out that always works — which is why point 9 isn't negotiable, and why I'd rather you never need it.

So it's public, it's in alpha, and we're building it out in the open. If it turns out to be useful to other people, wonderful. If it doesn't, I still end up with the thing I've wanted since 2000.

If you've read this far, there's a decent chance you've had the same argument with your own notes that I've been having with mine. Write a few. See what comes back.