Claude Code's auto memory makes your project less portable.
What is auto memory?
What is auto memory?
Around Feb 2026, Anthropic shipped auto memory to Claude Code. With it, Claude Code auto “remembers” your preferences and noteworthy info by writing notes into your home folder, at ~/.claude/projects/<project>/memory/.
The problem: those notes live outside your project folder.
Claude Code was designed to work within a project folder. An important best practice, and one I follow on every project: keep the project folder self-contained, version control it with Git, and push it to GitHub.
Two main benefits of this practice:
-
Your project is portable across machines. You can clone it on any machine, or work on it with Claude Code web, without loss of context.
-
Your project is portable across AI providers. When Claude Code is down, or you decide to switch, you can move to another provider such as Codex without loss of context.
With auto memory on, some important context is no longer self-contained in the project folder. When you change machines or switch AI providers, that context does not follow the project, and you will see drift in performance.
How to switch it off?
Add this to ~/.claude/settings.json:
{ “autoMemoryEnabled”: false }
Or prompt Claude Code to do it for you:
“Turn off auto memory for all projects globally.”
Two things to take note of:
-
The setting is per user, per machine. If you work across multiple machines, you need to do it on each one.
-
If you have already accumulated a lot of auto memory, chat with Claude Code to port those memories into your project’s CLAUDE.md before turning it off.
#AI #ClaudeCode #VibeCoding #AIAgents #BuildInPublic
If you are a non-techie who wants to learn to use Claude Code in the Terminal, I run a Foundations of Claude Code workshop.
Details: https://lnkd.in/e9r8kScw
If you are already familiar with Claude Code in the Terminal, I run an Advanced Claude Code workshop to deep dive into context engineering, skills and workflow patterns to run big and long-running tasks (a.k.a. Loop Engineering).
Details: https://lnkd.in/eR6WPQnC
Enjoyed this? Subscribe for more.
Practical insights on AI, growth, and independent learning. No spam.
More in Vibe Coding
Hitting your AI coding usage limit feels like reaching the climax of a drama series and having to...
You’re full of ideas, but suddenly on hold until next day.
UX/UI and naming matter more than capability for adoption.
Recently, everyone on LinkedIn is talking about learning Claude after the demo of Claude Cowork.
Claude Code and OpenAI Codex Do Track You
Recently, after hitting my Claude Code Max limit, I switched over to OpenAI Codex to continue my work.
We were promised autonomous AI agents. But got Workflow Automation 2.0 instead.
2025: The Year of AI Agents 😄
Two Choices for Handling Tech Debt in Vibe Coding
· Go full vibe: ignore tech debt, and when things inevitably break, spend a week fixing it.
"Guys, I’m under attack"
I came across this post where a founder shared how his SaaS got exploited right after he started sharing how he built his SaaS using Cursor.
Hitting your AI coding usage limit feels like reaching the climax of a drama series and having to...
You’re full of ideas, but suddenly on hold until next day.
Claude Code and OpenAI Codex Do Track You
Recently, after hitting my Claude Code Max limit, I switched over to OpenAI Codex to continue my work.
"Guys, I’m under attack"
I came across this post where a founder shared how his SaaS got exploited right after he started sharing how he built his SaaS using Cursor.
UX/UI and naming matter more than capability for adoption.
Recently, everyone on LinkedIn is talking about learning Claude after the demo of Claude Cowork.
We were promised autonomous AI agents. But got Workflow Automation 2.0 instead.
2025: The Year of AI Agents 😄
Two Choices for Handling Tech Debt in Vibe Coding
· Go full vibe: ignore tech debt, and when things inevitably break, spend a week fixing it.