Auto memory

Auto memory is the half of Claude Code’s memory system that you do not maintain. As it works, Claude saves things worth keeping — the actual build command, why a test is flaky, which service the staging environment points at — and carries them into later sessions.

Why it exists separately from CLAUDE.md

CLAUDE.md captures what you already knew to write down. Auto memory captures what was learned in the doing, which is usually the more expensive knowledge to reconstruct.

The two are complementary and are documented on the same page:

	CLAUDE.md	Auto memory
Written by	You	Claude
Lives in	Your repository	Claude Code’s own storage
Version controlled	Yes	No
Shared with the team	Yes, via the repo	No
Good for	Standards and intent	Discovered facts

Full comparison.

What it does not do

• It does not replace CLAUDE.md. Anything the team needs should be in the repository.
• It is not shared. Memory learned on your machine is yours.
• It is not a substitute for documentation. If a fact matters beyond Claude, write it down where humans will find it.

What this page does not claim

Storage location, retention, per-project scoping and how to inspect or edit what has been remembered are documented upstream and are not reproduced here.