A personal AI brain you own

It depends on what your work loses when the memory resets. If you use AI casually, the built-in memory in ChatGPT or Claude is probably enough: it holds preferences and a sketch of past chats, and the stakes of losing it are low.

The calculation changes when accumulated context is part of how you earn. A consultant’s client history, an operator’s decisions and exceptions, a writer’s voice — re-explaining that from scratch every session is a real cost, and so is discovering that years of it are locked inside one login you might cancel. That is the quiet form of lock-in: not a contract, just a memory you can’t take with you. People who hit it say the same thing — you don’t really own it.

The observation this page rests on: what you can read in plain text without logging in anywhere is yours; everything else is a feature. A personal AI brain is the yours column. For why the resets happen at all, see why AI forgets.

Less than people expect. A personal AI brain is not an archive of everything you have ever written. It is a small set of short files that carry what a capable assistant needs to work like it knows you:

• How I work. Your standards, your process, what a finished piece of work looks like, what you refuse to do.
• Decisions. The calls you already made and the reasoning: pricing, positioning, tools, boundaries. This is what lets a tool push back instead of agreeing with everything — it can cite your own decision back at you when a new idea contradicts it.
• Exceptions. Where your rules bend, and for whom. The part no generic assistant can guess.
• Voice.A few samples of writing that sounds like you, so drafts start in your register instead of a press release’s.
• Current work. One page per live project: where it stands, what is decided, what is open.

Each file is a page or two. Written once, updated when reality changes. Most people can draft the first version in an afternoon, and the structure is the same one context engineering applies to a whole business.

Yes, and this is the technical heart of the idea. Because the brain is markdown files rather than a vendor feature, every model can read it. The simplest move is manual: attach or paste the relevant files at the start of a session. It is unglamorous and it works today, in any tool.

The connected version uses the Model Context Protocol (MCP), an open standard for giving AI tools access to outside data. An MCP server sits in front of your folder and any compatible assistant — Claude, ChatGPT, Cursor — can query it directly, so you stop pasting and the tool starts reading. Running npx @br8n/mcp serves a folder to any MCP-compatible model, free, no account required.

Either way, the property that matters is the same: the memory is not inside any product. Switching models becomes a decision about capability and price, not about walking away from your own history.

Because for them, the accumulated context is the business. A consultant moves between client environments and rarely controls which AI tools each client allows, so memory tied to one vendor login breaks every time the environment changes. An operator running a small company holds pricing history, client exceptions, and process knowledge that no employer-provided tool will ever hold for them.

For that kind of work, a personal AI brain is the difference between an assistant that starts cold every engagement and one that arrives already briefed. It is owned, portable working memory in the most literal sense: your working knowledge, in files you own, wherever the work happens. Companies face the same question at a larger scale; that version is covered in owning your AI context.