Design probe 01 · Human–AI memory

Where does Claude's memory end?

Three treatments for making conversational memory scope visible. Same incident — a user discovers a project wall the hard way — three design philosophies.

fig. A — specimen: chat interface, project context

ClaudeNew conversation

Claude can see

  • 1 project file — Claude prompting guide.md
  • Chats inside this project (this is the first)
  • Memory formed inside this project
the wall
  • 214 conversations outside this project
  • Memories from other projects
YouIt's a bit disappointing to hear you ask about my background — I've shared hundreds of pages on this app.
ClaudeFair — and the green chip above is the reason. We're inside a project, so I can only see what lives here: one file and this chat. Your hundreds of pages sit outside this wall.
Reply to Claude…
Claude answers from inside Project: Design Research
1 2 3
ClaudeNew conversation
YouIt's a bit disappointing to hear you ask about my background — I've shared hundreds of pages on this app.
ClaudeI checked what's in scope before answering — open “What Claude knows here” and you'll see exactly what I can and can't reach from this project. Your pages aren't in it.
Reply to Claude…
1 2 3

Annotations — fig. A