Claude Memory: How the Feature Works (and What It Doesn't Do)
Claude can remember you now. After years of starting every conversation cold, Anthropic rolled...
Claude can remember you now. After years of starting every conversation cold, Anthropic rolled out memory in 2026 — and it's a real change in how Claude works. If you're trying to understand what the Claude memory tool actually does, how to turn it on, and whether it solves the "AI doesn't know me" problem, here's the straight version.
What is the Claude memory feature?
It's Claude's ability to retain information about you across separate conversations, instead of forgetting everything when a chat ends. When memory is on, Claude notices things worth remembering — a stated preference, a key fact about your work, your communication style — and stores them, then draws on them in future conversations without you repeating yourself.
This is a genuine shift. Before it rolled out, Claude was stateless by default: every conversation started from zero. Older guides saying "Claude doesn't remember anything between chats" are simply out of date now. Memory has been available on paid plans since late 2025 and rolled out more broadly in 2026, including an option for free users. (For comparison, OpenAI's memory works on a similar but separate model.)
How does Claude's memory work?
Automatically, and fairly transparently. You don't have to tell Claude to remember most things — when it detects something worth keeping, it creates a memory entry on its own, though you can also explicitly ask it to remember (or forget) something. When you start a new conversation, Claude reads relevant stored memories and factors them in.
A few details worth knowing:
- It's transparent. Claude is generally designed to tell you when it's drawing on stored memory, rather than using it silently.
- It's editable. You can view what's stored and delete or correct entries in Settings, and correct things in conversation.
- It's project-scoped where you want it. Projects can keep their own separate memory, isolated from your global picture.
- It synthesizes over time. Rather than just storing raw entries, Claude works your conversations into a usable picture of you, with processing that updates over roughly a day.
Exact menus, availability, and behavior vary by plan and are still evolving, so check your own Settings for what's active on your account.
Can Claude import memory from ChatGPT?
Yes — through a one-time import, not a live connection. Claude has a memory import tool (at claude.com/import-memory, or via Settings) that takes context exported from ChatGPT, Gemini, or Grok and turns it into Claude-native memory. You paste in a structured export from the other tool, Claude processes it (this takes time, not seconds), and your accumulated context seeds Claude's memory.
The important word is import. It's a migration, not a sync. New things you tell ChatGPT next week don't flow into Claude — you'd import again. It's designed to help you switch without starting from zero, not to keep two tools in lockstep.
How to turn on and manage Claude's memory
Enable and manage it in Claude's Settings (look for the Memory section). There you can turn it on, see what Claude has stored, delete or edit entries, and import memory from another tool. If you want a conversation that doesn't use or update memory, Claude offers ways to keep specific chats out of it. Because the exact layout shifts with updates, the Settings → Memory area is the place to look rather than any fixed menu path.
What doesn't Claude's memory do?
Three real limits:
- It's reactive. Memory is built from what you happen to discuss. It starts empty and fills in slowly, and it reflects your conversations rather than an accurate, structured profile drawn from your real work.
- It's Claude-only. Your Claude memory lives on Anthropic's servers and works inside Claude. Switch to ChatGPT, Cursor, or Gemini and none of it follows — every tool builds its own separate picture of you.
- It's personal, not shared. There's no built-in way for a team to build and share a common memory layer; each person's memory is their own.
For "remember what we discussed in Claude," the feature is good and getting better. For "every tool I use understands who I am," it falls short by design — because it was built to make Claude remember, not to make your context portable.
Memory vs a portable context layer
These solve different problems and work well together. Claude's memory handles continuity within Claude — recalling your past conversations and preferences. A context layer handles the foundational layer across every tool — a structured, current picture of who you are, served to Claude, ChatGPT, Cursor, and anything else through MCP.
Turn on Claude's memory; it's worth it. But if you're using more than one AI tool and tired of each one knowing a different slice of you, the fix isn't a better single-tool memory — it's context that lives outside any one tool and travels to all of them.
→ The deeper comparison: Unabyss vs Built-In AI Memory
→ Does Claude remember across chats: Does Claude Remember You?
→ Give every tool a current picture of you — Unabyss →
