Does Claude remember previous conversations?

This is the most common question people hit in their first week of serious Claude use, and the answer changed meaningfully over the past year, so most advice you'll find is stale. Here's the current picture, and how I actually handle it while building businesses inside Claude every week.

What does Claude's memory feature do?

Claude has an opt-in memory feature that summarizes your conversations and carries key context into new ones. It's transparent by design: you can open settings and see what Claude remembers, edit it, tell Claude to focus on or ignore specific things, or switch it off. Projects get their own separate memory space, so what Claude learns about one workstream doesn't leak into another. There's an incognito mode for conversations you want left out entirely.

Used well, memory removes the re-explaining tax. Claude knows what you're building, how you like to work, and what you decided last week, without you pasting a briefing into every chat.

What about searching past chats?

Separately from memory, Claude on paid plans can search your conversation history on request. Ask "find our discussion about the pricing change" and it retrieves the relevant chat and pulls the context in. It's genuinely useful for the "I know we talked about this somewhere" moment.

The gap: memory is context, not a record

Here's the distinction that matters. Memory answers the question "does Claude know me?" It does not answer "where is that decision written down?" You can't open Claude's memory in a text editor. You can't grep it, commit it next to your code, print it, or take it with you if you switch tools. It's context that makes conversations better, held inside a product, on a schedule and format the product controls.

Your decisions, tasks, and numbers deserve a stronger guarantee than that: a file, in plain text, that you own. Memory and a file aren't competitors; they're different layers. One makes Claude smarter about you. The other is your actual record.

How I run it

Memory on, search available, and a capture command for anything that matters: mid-conversation, I type /jot and the decision or task gets filed by project into one Markdown file, jotthread.md, that lives with my project and syncs anywhere. That's JotThread, the skill I built for exactly this split. If you'd rather build the habit with free methods first, this guide covers all three approaches, and if you're organizing multiple workstreams, start with Projects.

Common questions