Disable Auto-Memory to Stop Unexpected MEMORY.md Writes

Claude Code's auto-memory feature writes MEMORY.md and topic-specific files to ~/.claude/projects/{path-hash}/memory/ whenever it detects something worth remembering. That sounds useful — but it can consume tokens unexpectedly, cause sessions to behave inconsistently, and accumulate stale context that quietly misdirects Claude over time.

You can turn it off with a single line in your project or global settings.json:

{
  "autoMemoryEnabled": false
}

For a project-level setting, put this in .claude/settings.json in your repo root. For a global default, add it to ~/.claude/settings.json.

Setting autoMemoryEnabled to false stops new memory writes but preserves everything already captured. You won't lose existing context — and you can re-enable the feature later without any data loss.

It's worth opening ~/.claude/projects/{path-hash}/memory/ and reviewing what's already there. Stale or inaccurate entries can subtly influence Claude's behaviour across sessions without you realising it. Editing or deleting individual files is completely safe.

If you want to keep auto-memory on but stay in control, you can also toggle it interactively mid-session using the /memory command inside Claude Code.

Auto-memory is only helpful when it's accurate — review it before it drifts.

via @coreyepstein