Why ContextHouse

Context is king. Why keep it yourself?

If the front page left you with questions, this is the explanation I’d give you across a table: what context actually is for an AI, why your answers get better with your own files in front of it, why it matters whose folder they live in, what the whole thing honestly costs, and who shouldn’t bother.

What “context” means for an AI

An AI assistant doesn’t remember you the way a colleague does. When it answers, it works from what is in front of it right then: your question, the conversation so far, and whatever instructions or files are attached. That is its context, and it is the whole of what the model knows about you in that moment. Close the chat, open a new one, and it starts from nothing again.

So most people spend the first minutes of every serious session re-explaining themselves. What the project is, what was decided last month, what the client is like, which of the three PDFs is the current contract. Some of it gets typed, most of it gets skipped, and the answer comes back wrong in the one detail that mattered.

The makers of the models say this themselves. Anthropic’s own guidance for getting good work out of Claude is to treat it like a brilliant but very new employee who knows nothing about your situation, and to put the context behind the task in front of it; OpenAI’s advice comes down to giving the model any additional information it might need.

ContextHouse turns that into a habit. Your context lives as plain files in one folder: the week on one page (Now.md), the projects with their plans and logs (Projects/), the decisions with their reasons (Decisions/), the people, the paperwork. You connect your AI to the folder when you start a chat. From then on, the instruction files in the house tell it which files to read before it answers, and where to file what comes up.

What actually gets better

Mostly, you stop starting from zero. Today your information probably lives across chat histories, a notes app or three, screenshots, downloads and your head, and none of it connects. So every morning you brief your AI from scratch, the question you settled in March gets re-argued in July because nobody wrote the answer down, and the insurer’s PDF is somewhere in Downloads under a name like scan_final_2.pdf.

With the house, the same days run differently. The AI reads your week file before it says anything, so the morning briefing disappears. The March decision sits written in the decisions folder, and the AI checks that folder before advising you, so July stays quiet. The PDF was swept into the insurance folder under a clean dated name the day it arrived, together with a note on what you decided and why. And when a chat dies in the middle of a project, you lose nothing, because the project’s state lives in its files rather than in the window you just closed. You open a new chat, say “continue”, and it picks up at the written next step.

Give it a year, and it compounds

Here is one thread the way it actually runs:

  1. Week 1. You talk through switching your legal insurance with the AI. The policy PDF gets swept into your files under a clean dated name; what you decided, and why, is written down.
  2. Week 3. A dispute comes up. You say “the thing with my legal insurance” and the AI already has the policy, the coverage, and your own reasoning in front of it.
  3. Months in. The next question, raise the deductible or not, starts from everything the first two conversations left behind instead of from zero.
  4. A year on. The renewal window comes close, and the cancel-by date surfaces in your week view, because the policy was filed once. Nothing runs in the background; the date is simply there when you open the house.

A year in, the AI argues from the policy, the reasoning and the dates you filed yourself, none of which it had on day one.

Why the folder should be yours

The AI companies know that context is the whole game, which is why every app now has a memory feature. Switching it on helps, and it is also a different thing from what I’m describing. Built-in memory is a summary the app extracts from your chats: that you have a dog, that you prefer short answers. You can view it and delete from it, but you can’t really organize it, it lives inside that one company’s app, and moving it to the next app is a copy-paste job at best. It was never meant to hold your lease, your project plans, or the reasoning behind a decision.

Keeping your context in a folder you own is the strategic move, and it pays in three ways. Your AI argues from the real material instead of a summary of you. You can change assistants without losing a year of context: the folder is plain Markdown, the same files work with Claude now and with ChatGPT if you switch next year, and when local models are ready, with them too, on your own computer and off the big providers entirely. And you get a routine that doesn’t depend on anyone’s app: everything new goes into one inbox, the AI files it on “sweep”, and once a week the Steward’s fifteen minutes keep the house current. The folder stays on your disk whatever app you use next, and the next assistant reads it under the rules written in CLAUDE.md.

The honest costs

Three things should be said plainly before you spend money on this idea, mine or anyone else’s:

Who should skip it

If you enjoy building systems, the free kits on GitHub are a good weekend, and you may want yours handmade. If your life genuinely fits in one notes app and that’s working, keep it. A house for your context earns its keep when the volume of projects, paperwork and decisions has outgrown what one app and one head comfortably hold.

If you’ve heard “second brain”

That is the older name for this idea, and ContextHouse is a version of it. People have always kept memory outside their heads: notebooks, diaries, filing cabinets, and in the famous case of the sociologist Niklas Luhmann, a wooden box of ninety thousand paper notes that referenced each other, which he wrote about in 1981. The files in the house link to each other too, which is why people describe the result as a personal Wikipedia. What changed recently is that an AI can now read and write that folder with you, and that turns it from an archive you maintain into something closer to a colleague who has known you for years and takes notes. If you want the original, his card box is digitized at the Luhmann archive.

What the finished version looks like

ContextHouse is my version of this idea, already built: 52 plain files, a guided setup that moves you in by conversation, and a caretaker that keeps the house true in fifteen minutes a week. If the case above lands, see exactly what you get, or inspect all 52 files before paying anything.

Get ContextHouse · €19.81

Written by Stefanos, who runs his company on this exact system.