← All writing
05 Jul 2026

Your AI context should not live in your AI vendor

A portability stance, not a privacy one: keep your project's history, decisions, and state in three plain files you own, and any AI tool can pick up exactly where the last one left off.

Your AI context should not live in your AI vendor.

That’s not a privacy stance, it’s a portability one. If the only place your project’s history, decisions, and current state exist is inside a chat thread, you’ve built your workflow on top of someone else’s product roadmap.

I tested this the hard way this week. Moved my entire work tree out of a cloud-sync folder (it was quietly corrupting Python venvs and symlinks) into a plain local projects folder, and switched AI subscriptions on top of it. Both changes, same afternoon, under four hours. Six projects, one hub.

The structure is boring on purpose

One folder holds all projects. Each project gets three files: CLAUDE.md for what it is and how it’s built, STATUS.md for where it stands right now, DEADLINES.md for what’s time-sensitive. Anything crossing project lines gets a short dated note instead of living in someone’s memory.

projects/
  hub/
    CLAUDE.md, STATUS.md, DEADLINES.md, briefs/
  project-a/
    CLAUDE.md, STATUS.md, DEADLINES.md
  project-b/
    CLAUDE.md, STATUS.md, DEADLINES.md

When I opened a fresh chat under the new subscription, it read those files and picked up exactly where the old one left off. Not a reconstructed summary. The actual state.

Setting this up takes an hour, not a quarter

  1. Pick one folder as home for all your projects. Not synced storage, not a vendor’s workspace. Just a folder on disk.

  2. Inside each project, add three files: CLAUDE.md, STATUS.md, DEADLINES.md.

  3. Update STATUS.md before you close a session. Two sentences is enough.

  4. When something crosses project boundaries, write a one-page dated brief instead of relying on memory.

  5. Point any AI tool at the folder. Read the three files first, then work.

That’s the whole trick, and it isn’t clever. It’s just refusing to let a vendor’s chat history become your system of record for how a project works.

Models will keep changing. Subscriptions will keep changing. None of them are your memory layer, they’re compute you rent by the token. The context that makes an AI useful to your actual work belongs in files you own, under version control, readable by anything.

Move the files, not the chat.

Source post: Post | LinkedIn