Every word the AI reads costs something.
Time, money and attention: a model handed a hundred notes to answer one question takes longer, costs more and gets more wrong. A shared memory built well is not just a place to keep things: it is a way of handing them to the AI in the right order, and in the right amount.
The context is the window
Everything the AI knows in a conversation comes through it.
A model does not «know» your things: it reads them, every time, inside the conversation’s context window. What you put in there is paid for by whoever pays for the model, in tokens — and paid for again in the quality of the answer: the more noise there is, the easier it is for the model to lose the thread or invent.
The naive way to give an AI memory is to dump everything on it. The right way is to give it first the things that always hold, then the ones that hold for this job, then the rest if needed. And to tell it what it has not read.
How Cobrain serves it
In layers, with a header saying what is inside.
Rules first, then the card, then the notes
A folder with an index card is a unit: a client, a project, whatever you follow. When your AI opens a unit it receives, in full, the rules for the whole brain, that unit’s card, its instructions and the latest decisions. Only then the notes in the branch. It is the same order you would use with a new colleague.
Whole lines, never fragments
Search returns the lines that matter, whole, not pieces cut in half. One cut fragment once turned «Alessandro’s sister is called Alessia» into «Alessandro is called Alessia». Since then lines are handed over whole, and the AI is told to read the note before answering.
A stated ceiling
When it loads a large branch, Cobrain stops at a limit and says so at the top: the AI knows it has not read everything, and can ask for the rest instead of answering as if it had.
The brain is never dumped whole
A brain with six hundred notes never passes into a conversation in full. The AI searches, finds the few notes that matter, reads them. That is the context — not everything.
What it means for you
Better answers, shorter conversations, smaller bills.
- Less to explain again. The brain’s rules and the client’s card arrive on their own, at the start, every time.
- Fewer tokens. On a pay-as-you-go plan, like the APIs or Claude Code, useless context is a line on the bill. Here there is none.
- Less invention. A model that reads whole lines and knows what it has not read invents less than one that reads fragments and believes it read everything.
- The same for every AI. Claude, ChatGPT and Claude Code receive the same context, in the same order.
How to organise the brain to spend less
Three habits that pay.
- An index card for everything you follow. A client, a project, a person: a folder with the card at the top. The AI gets there with one command and knows where to start.
- Decisions in a single file, appended, with the date. It is the first thing the AI reads after the card, and the thing that stops settled arguments being reopened.
- The general rules at the root. How you want the AI to write, what it must not do, which names to use: one note at the root, and it holds for every unit and every AI.
The tools the AI uses to do all this are documented in full.
Give your AI the right context.
An index card, the decisions, the rules. Then you ask about the client and it starts out already briefed.
