How to Make Claude Remember Your Projects and Preferences

Why Claude starts every conversation from zero

Claude is brilliant in the moment and forgetful by design. Every new conversation begins with a blank slate, so it has no idea what you are building, who your teammates are, or how you like your emails written. You end up re-explaining the same context again and again.

This is not a Claude flaw, it is how almost every chat assistant works. The model does not carry your life between sessions. If you want Claude memory that actually sticks, you have to give it a place to read from. We dug into the root cause in why your AI keeps forgetting you.

Step 1: Put your projects, people, and preferences in one folder

Make a single folder on your computer and fill it with short plain text notes, one per thing you care about. A file for each active project. A file for each important person. A file for your preferences (tone, formatting, the tools you use). Write them in Markdown, which is just text with a few headings and bullet points, nothing technical.

Keep each note small and specific. Project Atlas gets its own file with the status, the deadline, and the people involved. Six months from now, that folder is a real record of your work, readable by you and by any AI. This is the plain-files idea behind a second brain. If you already use Obsidian, good news: an Obsidian vault is already a folder of plain files, so there is nothing to convert.

Step 2: Connect Claude to the folder

Claude can read files on your computer through connected apps, using an open standard called MCP (Model Context Protocol). You do not need to memorize the acronym. Think of it as giving Claude a library card for one specific folder.

In Claude Desktop, you add a small connector that points at your notes folder. Claude Code, the terminal version, connects the same way. Once it is set up, Claude can open and search those files during a chat. It is a one-time step, and the better tools ship a plain-language guide for each app so you are not editing config files by hand. If you want the same trick for other assistants, see how to connect a folder to any AI.

Step 3: Ask, and check where the answer came from

Now just talk to Claude like normal. What is the status of Project Atlas? How does Sara like her updates? Draft a reply in my usual tone. Claude pulls the answer from your notes instead of guessing, so the reply is grounded in your real work, not a plausible invention.

The detail that matters most: a good memory setup shows you which file each answer came from. If Claude says the deadline is Friday, you can open the note that says so. That is the difference between an assistant you can trust and one you have to double-check every time. A memory that cannot cite its source is just a confident guess.

But doesn't Claude already have a memory feature?

It does, and it is genuinely handy: zero setup, free, and it quietly remembers a few things about you across chats. The catch is where that memory lives. It sits inside Claude, in a form you cannot read, edit, or take with you. Switch to ChatGPT or Cursor for a task and none of it comes along. Cancel your plan and it is gone.

Built-in memory also stays locked to one vendor. The folder approach flips that: your notes are yours, in plain files, and every AI reads the same source. We compare the two in detail in Callosium vs built-in memory, and we looked at who actually controls it in your AI memory has an owner.

Teach Claude once, and every other AI knows it too

This is where a folder beats a built-in feature. Because the memory is a folder on your machine, ChatGPT, Cursor, and Claude can all read the same notes. Update your preferences once and every assistant is current. You are not maintaining five separate memories, you are maintaining one.

This is exactly what Callosium is built to do: a local-first shared memory that any AI connects to. It adds two things a plain folder cannot. Each AI only sees the folders you grant it, enforced by the software rather than by a promise. And every change is stamped with who made it, you or a specific AI, so you always know where a note came from. It runs on your machine, answered about 19 of every 20 questions on a fresh 15,000-question test, and returns most answers in around 28 milliseconds. You can read how to build a second brain every AI can read to set the folder up the durable way.

Why plain files beat rented memory

Here is the real risk with memory that lives inside someone else's product: it can be switched off. This year, one of the best-funded AI memory apps was acquired and began shutting down, with apps discontinued, some regions cut off, and accounts scheduled for deletion. People who had trusted it with years of context were suddenly locked out.

Plain files on your own computer cannot be sunset. The free, local-first approach never uploads your notes anywhere; the AI reads them where they sit, and nothing is sold. If you ever want the same notes on your phone and laptop, that stays your choice through your own cloud, and the core stays free forever. Early access to Callosium opens 4 August 2026 at callosium.com.

Frequently asked questions