Skip to content

AI Compass / Knowledge cards

Memory in three tiers

what Claude knows about you and your projects

Checked Prompting Everyday

A chat knows only what is in its history. Jack Roberts therefore describes a memory in three tiers: a short instruction about you that applies everywhere, a folder per work area with its own CLAUDE.md and notes, and an archive for history and expertise. His goal: a new chat answers just as well right away as an old one. Claude Code also comes with its own automatic memory, in which Claude itself stores notes about you and your project.

What it is for

You do not have to explain yourself, your goals and your projects anew in every chat. You can find last week’s decisions again.

Steps

  1. Group your own activities into 6 to 8 areas, no more.
  2. Tier 1: a short instruction with name, role, goals, answer style, tools and taboos, under 200 words. In Claude Code, according to the docs, it belongs in ~/.claude/CLAUDE.md and then applies to all projects. The video shows the Claude app.
  3. Tier 2: for each area, a folder with a short CLAUDE.md and a memory/ subfolder for purpose, technology, decisions made and an overview of what is where. Always start in the right folder, because according to the docs Claude Code loads the CLAUDE.md of the folder you work in.
  4. Create a profile as its own file: current strategy, focus, decisions. It is read in every session. You can include it in the CLAUDE.md with @path/to/file (according to the docs).
  5. Tier 3, the archive: notes as Markdown files, for example in Obsidian if you want to read them yourself. A vector search (search by meaning instead of by keyword) pays off only with a very large amount of text. The area’s CLAUDE.md says when to consult the archive.
  6. At the end of every session, have a summary with date, decisions and next steps written into the archive. Jack Roberts uses a custom command “/wrap-up” for this. The video does not show how it is built.
  7. Know the automatic memory: According to the docs, Claude Code stores notes about your preferences, your corrections and the project status under ~/.claude/projects/<project>/memory/ and loads the first 200 lines or 25 KB of the file MEMORY.md at the start of every session. With /memory you can see it and switch it on or off. Read it now and then and delete anything wrong.

Limits

Sources

Status

Checked 17 October 2026 · Maturity new (concept available, built-in tools described in the docs; effect not measured)

Want to try this with your team? That’s what the AI Lab is for. See the AI Lab