
SimplifyUX Notes
A notes app that keeps every note as a plain file, so people and AI agents read the same thing. Part of the SimplifyUX suite, the software I'm building for my own company.
SimplifyUX Notes is where the SimplifyUX suite keeps what it knows. The suite’s other products are the design tool, the console that runs the websites and the task board that plans the work. It is the second product in the suite I’m building for my own company, SimplifyUX, and the one with the simplest rule: every note is a plain file. A person reads it in a comfortable editor. An AI agent reads it the way it reads any file, with nothing in between. The screens here are rebuilt around invented notes for an example project, a ferry timetable app, so nothing in them is real.
What was broken
My notes lived inside a commercial app that stored them its own way. That was fine while only I read them. Once AI agents started doing most of the building in my work, they needed to read those notes too: the decisions, the research, the reasons. Every route in went through the app, and every route had to be set up and trusted.
The deeper problem was ownership. Notes kept in someone else’s format are only as durable as that company and that format. I wanted notes I could open in ten years with any text editor, back up like any other file, and hand to an agent without building anything.
How I worked it out
I set the rules and Claude wrote the code, as with the rest of the suite. The first rule decided most of the design: the files come first and the app second. There is no database. The folder of Markdown files is the whole of the data, and the app is one way of looking at it.
That rule made three things hard, and each became a design decision:
- A rich editor that never damages the file. Rich editors like to rewrite what they open. This one keeps the block of fields at the top of each note exactly as it was, byte for byte. A test in a real browser proves it on every build.
- Search without an index. An index is a second copy of the notes that can fall out of step. Search reads the files themselves, so what it finds is always what is there.
- No AI inside the app. I kept it out on purpose. Agents read the files directly, so the app never needs to send a note anywhere, and nobody has to trust a feature to keep a secret.
- 01CaptureStart blank or from one of five templates.
- 02OrganiseFolders and tags, kept in the file itself.
- 03FindSearch reads the files, so nothing goes stale.
- 04ReadA calm reader that can read the note aloud.
- 05CiteReferences in IEEE style and a note on how AI was used.
What I designed
The editor looks like a page, not a form. Headings are set in a serif and the interface around them in a plain sans. A single toggle switches between the rich view and the file itself, frontmatter included. It saves on its own 700 milliseconds after you stop typing, and a small status says so in words: saved, saving, unsaved changes or save failed.
Search shows each match with its file path, because the path is real: it is where the file lives on disk. The reader drops the interface away and sets the note in a large serif, with a player that reads it aloud a paragraph at a time. Templates give a note its shape before the first word. The academic one adds references in IEEE style and a declaration of how AI was used. The record of AI help travels with the work instead of being rebuilt at the end.
Accessibility is a setting, not an afterthought. Theme, text size, contrast, font, motion and the strength of the focus ring all change on the device. The app also installs to a phone and keeps working offline for reading. One colour choice shows the care involved: the brand red never sets small text, because it misses the 4.5 to 1 contrast that small text needs.
Every way into a note is the same file, which is the point of the whole design:
| Reader | Way in | What they see |
|---|---|---|
| A person at a desk | The app in a browser | The rich editor, or the file in Markdown |
| A person on a phone | The installed app, online or offline | The reader, with read-aloud |
| An AI agent | The file on disk, or a small JSON interface | The same Markdown, fields and all |
| Me, in ten years | Any text editor | The same Markdown, still readable |
What changed
The first version shipped on 30 June 2026, the day the work began: 18 commits from an empty folder to a working editor. The commercial app it replaced is no longer installed. Since then the work has been small and steady, which is what a notes app should need.
The guard rails are tested, not promised. Every file path is kept inside the notes folder: no climbing out with two dots, no absolute paths, no hidden characters, no links that point elsewhere. A single note is capped at one megabyte. The browser is held to a strict security policy, and a test in a real browser fails on any breach. Sixteen tests drive the whole app in a real browser. They run in four browser setups, 64 runs on every build, including automated accessibility scans of every page.
On 13 October 2026 one of those tests earned its keep. An update left two copies of the editor’s core library side by side, and typing quietly broke. Only the browser test for the editor caught it. The same day, every known security advisory against the app’s dependencies was cleared: 20 findings before, none after.
Commits by month
What I learned
The best decision in this product was a refusal. Keeping AI out of the app made it simpler, safer and more useful to AI at the same time, because the agents never needed the app at all. They needed the files, and the files were already there.
Plain formats outlast clever ones. A database would have made search faster to build and the notes harder to keep. Choosing the file as the source of truth made a few things harder for me and everything easier for whoever reads the notes next.
And small tests in a real browser catch what reading the code cannot. The broken editor in October looked fine in every file that changed. It only failed when a browser tried to type.









