A personal OS built with Claude Code
Claude Code is a coding tool, and this is not a coding use for it. The reason it turns out to be the right foundation for a personal operating system has nothing to do with code and everything to do with one capability that chat interfaces do not have.
The capability that matters
Claude Code can read and write files in a folder on your own machine, directly, without you pasting anything. That sounds mundane. It is the entire difference between a system that maintains itself and a system where you are the integration layer.
In a chat window, the assistant can reason about your week beautifully and then hand you text. You copy that text somewhere. You are the part that moves information from the conversation into the record, every time, forever. That is precisely the maintenance tax that kills hand-built systems, and a chat interface does not remove it. It just makes the thinking upstream of it better.
With direct file access, the conversation and the record are the same event. You say a project closed; the project file changes. You talk through a decision; the reasoning lands in the decision log. Nothing was copied, so nothing was forgotten.
What each layer looks like in practice
| Layer | How it is built |
|---|---|
| Filing | A folder of markdown files, plus a standing instruction file the assistant reads at the start of every session that says where things go |
| Routines | Each routine written once as a skill, invoked by name. The weekly review is a file, not a habit you re-explain. |
| Acting ahead | Scheduled or on-demand passes that prepare reversible work and leave it for approval rather than sending anything |
| Improving | Corrections written back into the instruction files, so the same mistake is structurally unavailable next time |
The second row is the one people underestimate. Once a routine is a file rather than a prompt you retype, it stops degrading. You improve it once and every future run inherits the improvement, which is the opposite of how a chat-based workflow behaves.
What it costs you
Honestly: a terminal. Claude Code runs in one, and if the sight of a command line is genuinely unpleasant to you, that is a real cost and worth weighing rather than dismissing. What it does not cost you is programming. You will not write code to build this, and the setup is closer to describing how you want your files organised than to development work.
The test worth applying to any tool in this category: can it change a file on your disk without you doing the changing? If not, you are still the one keeping the system alive, whatever else it does well.
How it compares to the alternatives
Claude Projects holds context across a conversation but does not touch your filesystem, which is the distinction drawn out in Claude Projects versus Claude Code. Notion AI acts inside a workspace it owns rather than files you own, covered in Notion AI versus Claude Code. Obsidian gives you the files and no native intelligence, so every AI feature is a plugin you wire yourself, which is the trade in Obsidian versus Notion AI.
None of that makes those tools wrong. It makes them different halves of the same thing. The reason this site keeps returning to file ownership plus intelligence together is that most products give you one and quietly cost you the other, which is the argument in the AI second brain comparison.
Where to start
Start with the folder, not the tooling. A structure worth copying is at the template, and the order to assemble the layers in is at how to build a personal operating system. The tool is the last decision, not the first, and this page exists mainly because the tool question is the one people search for first.
Built hands-on, prompt by prompt
The book builds the whole system in Claude Code with every prompt printed on the page. Nothing to download, and no code to write.
Get the bookWhat it looks like at company scale
The same file-first approach running on a company's email and meetings, with the acting-ahead layer switched on. Try the demo.
See the demo