Yesterday I opened Claude Code to continue working on the same task I had been working on for a few days. But, the problem was I couldn’t resume the session because I had to recreate the container where I was using Claude Code. Like I expected, it asked me to explain the architecture again.

Claude sessions don’t actually persist memory across sessions. Every session starts stateless unless context is explicitly reloaded. It’s reconstructed by rereading project files each session.

Claude forgets sessions. Your filesystem doesn’t.

Claude Code memory behaves like persistence, but technically it’s deterministic file ingestion.

When Claude Code starts, it scans the project, and rebuilds context from whatever it finds. Technically it’s just deterministic reloading of structured notes.

So I added to claude.md:

## Memory Management

When something important happens — architecture decisions, bug fixes, gotchas —
append it to .claude/memory.md

Keep entries short: date, what, why.
Read this file at the start of every session.

And then: .claude/memory.md entries look like this:

- 2026-02-12: Auth service depends on FEATURE_X flag — removing breaks staging
- 2026-02-13: Containers must use internal API URL, not localhost

Whenever a session starts, Claude reads the file and I get the Context added back again. After using a few times, I started to feel like Claude has got a memory.

But technically it’s repeatable context loading from disk.