You explain your writing style to Claude in a normal chat, then open a Project to draft the actual piece, and Claude acts like it has never met you. That is not a bug. Claude runs two memories that do not talk to each other: a global memory that synthesizes your standalone chats, and a separate, walled-off memory inside every Project. Knowing which one you are in decides whether Claude remembers anything at all.
Most coverage frames this as Claude being forgetful inside Projects. It has it backwards. The wall is deliberate, and for some work it is the most useful thing Claude's memory does. The real question is not why the two are separate. It is when you actually want one brain instead of many, and what to do when Claude refuses to give you that.
Claude has two memories, and they never share notes
Claude's global memory works by synthesis, not recording. As of June 2026, Anthropic's help center states that Claude "will automatically summarize your conversations and create a synthesis of key insights across your chat history (not including chats in projects)," and that "this synthesis is updated every 24 hours and provides context for every new standalone conversation." So when you start a fresh chat outside a Project, Claude pulls from a rolling daily digest of everything you have discussed in other standalone chats.
Two consequences fall out of the 24-hour cadence. A detail you mentioned an hour ago in another standalone chat may not surface yet, because the synthesis has not refreshed. And the memory is a summary of insights, not a transcript, so Claude carries the gist of your past chats rather than their exact words. That is closer to how a colleague remembers you than how a search index does.
Here is the part most write-ups skip. You do not have to wait out the 24-hour clock at all. Tell Claude directly to remember something inside any standalone chat and it writes that fact immediately, applying to your very next conversation, no synthesis cycle required. Anthropic caps these manual notes (Claude stores a limited set of short, direct edits rather than an unbounded log), so they are best spent on the handful of facts you never want it to forget. The daily digest handles the patterns; the direct note handles the one thing you need locked in now.
This shapes what you can expect Claude to recall on its own. The synthesis favors durable patterns over one-off facts: your preferred tone, the languages you code in, the constraints you keep repeating. A single throwaway number you mentioned once is less likely to make the cut than a preference you have voiced across many chats. Treat global memory as a sense of who you are, not a vault for specific values. When precision matters, restate the fact or pin it with a direct note rather than assuming the digest kept it.
Claude's memory is a daily synthesis of your standalone chats. Projects sit outside it entirely, each with its own separate brain.
Every Project gets its own walled-off memory
Projects do not borrow from global memory, and they do not pool with each other. Anthropic is explicit: "Each project has its own separate memory space and dedicated project summary, so the context within each of your projects is focused, relevant, and separate from other projects or non-project chats." In the launch announcement the company puts it more plainly: "If you use projects, Claude creates a separate memory for each project."
Anthropic frames the isolation as a feature, not a limitation. Its stated reasoning: "This ensures that your product launch planning stays separate from client work, and confidential discussions remain separate from general operations." A Project about Client A never bleeds context into a Project about Client B, and neither one leaks into the casual chat where you asked Claude to explain a recipe. Each space stays focused on its own work.
Why the wall helps more often than it hurts
Separation is what makes a Project trustworthy for sensitive work. If you discuss an unannounced product inside a Project, you do not want that leaking into a standalone chat where a colleague borrows your laptop, or bleeding into a different client's Project. The wall keeps each context clean. It also keeps Claude sharp: a Project summary built only from one body of work gives better answers than a memory diluted by every unrelated chat you have ever had.
There is a privacy angle too. Confidential strategy lives in its own room with the door shut. Nothing from that room joins the global synthesis that flavors your everyday chats. For regulated or competitive work, that boundary is the whole point, and a single shared memory would quietly undermine it.
The wall also makes Claude's answers more accurate inside a Project. A summary distilled from one focused body of work points Claude at the right vocabulary and assumptions. If your Client A Project pulled in stray context from a Client B chat, you would get subtle contamination: the wrong product name, a constraint that does not apply, a tone borrowed from the wrong audience. Isolation removes that failure mode. The narrower the memory, the more reliably it fits the task in front of it.
If a piece of work is confidential or client-specific, do it inside a Project on purpose. The walled memory is the safest place for it.
When the wall is exactly what frustrates you
The same isolation that protects sensitive work blocks the unified-memory dream. You taught Claude your tone, your stack, your constraints across months of standalone chats. Open a Project to actually build something, and none of it follows you in. You re-paste the context. Then you start a second Project and re-paste it again. The thing you wanted, one Claude that knows you everywhere, is precisely what the architecture forbids.
It cuts the other way too. Insights you developed inside a Project, the sharpest thinking you have done with Claude, never reach global memory or any other Project. Your best context is trapped in the room where it was created. Anthropic's design optimizes for compartmentalized teams, not for an individual who wants one continuous relationship with the assistant across every surface.
The friction shows up most for people who live across many Projects. A freelancer with one Project per client, a researcher with one per topic, a developer with one per repository: each of them maintains the same baseline context in every room, by hand, forever. The wall that protects a confidential launch is the same wall that makes you the human glue between a dozen Claude memories. There is no built-in way to say "this preference applies everywhere" once and have it stick across the divide.
| Dimension | Global memory | Project memory |
|---|---|---|
| What it remembers | Synthesized key insights from your standalone chats | A dedicated summary built only from that one Project's chats |
| Scope | Cross-chat, but only non-Project conversations | Single Project, sealed off from everything else |
| Sync frequency | Synthesis updated every 24 hours; direct notes apply at once | Per-Project summary scoped to that Project's activity |
| Isolation | Shared across all your standalone chats | Walled off from global memory and other Projects |
| Controls | Pause, reset, incognito chats | Lives and dies with the Project; not in global memory |
| Best use | Ongoing personal context you want everywhere casual | Confidential, client-specific, or focused bodies of work |
The workarounds people use, and where they break
Since the wall will not move, people reach for manual bridges. The most common one is a context block: a short paragraph of your preferences and constraints that you paste into the top of each Project. It works, but it ages. Update your stack or change your tone, and you now have to find and edit that block in every Project, or live with stale instructions in the ones you forget.
Some people keep the canonical version of that block in a single standalone chat and copy from it. Others store it in a Project's instructions field so it loads automatically inside that one Project. Both are real improvements, and both confirm the underlying gap: there is no native shared memory, so you are maintaining a copy by hand. The more Projects you run, the more copies drift out of sync, and the more time you spend being your own sync engine.
The controls you actually have over each
Global memory comes with three switches. Pausing tells Claude to keep its existing memory but stop using it or making new memories. Resetting permanently deletes all memories, including project memories, and Anthropic warns the action cannot be undone. Incognito chats give you a per-conversation escape hatch: Anthropic describes incognito as a mode where Claude won't remember your chats, so they are not saved to memory or your chat history.
One detail to weigh before you hit reset: it does not spare your Projects. The same button that wipes global memory also clears every Project's memory, and there is no undo. Project memory is otherwise governed by the boundary itself. There is no toggle that lets a Project read your global memory, and no setting that merges two Projects into one shared brain. The control is structural: you decide what goes inside which Project, and Claude keeps each one separate for you. Memory from chat history is available across all plans: free, Pro, Max, Team, and Enterprise.
How to decide: one brain or many?
- Want Claude to know your style and preferences everywhere casual? Stay in standalone chats and let the daily synthesis carry context.
- Need a fact locked in right now? Tell Claude to remember it directly in a standalone chat; that note applies to your next conversation without waiting for the 24-hour refresh.
- Doing confidential or client-specific work? Use a Project on purpose, so the context cannot leak into your global memory or another client's space.
- Need the same context in two Projects? Accept that you will set it up in each one. The wall does not move.
- Want a truly fresh conversation with no memory at all? Use an incognito chat outside any Project.
When you want one brain across more than Claude
Claude's split is a sensible answer to one problem: keeping work compartmentalized inside Claude. It does nothing for a different problem, which is that your context is now scattered across Claude's global memory, each separate Project, and whatever ChatGPT and Gemini happen to remember on their own terms. Every tool, and every Project inside a tool, holds a partial copy of you. None of them is portable.
MemX is built for that gap. It is a single memory layer you own and carry across tools, so the same context follows you instead of fragmenting per app and per Project. You decide what it holds and where it applies, rather than re-pasting your brief into the next walled room. It is private by architecture: per-user isolation, encryption at rest, and your memory is never used for training. Claude's wall stays useful for confidential work; MemX answers the case where you wanted continuity and the wall took it away.
Frequently asked questions
01Does Claude remember conversations inside Projects?
Yes, but separately. Each Project has its own memory space and dedicated summary. That memory is walled off from your global chat memory and from every other Project, so context inside one Project never carries into another or into standalone chats.
02How often does Claude's memory update?
Claude's global synthesis of your standalone chats updates every 24 hours. But you can skip the wait: tell Claude directly to remember something and that note applies to your next conversation immediately, no daily refresh required.
03Does Project memory share with global memory?
No. Claude's global synthesis explicitly excludes chats inside Projects, and Projects do not read from global memory. The two are separate by design, which keeps confidential or client-specific Project work isolated from your everyday chats.
04Can I turn off Claude's memory?
Yes. You can pause memory so Claude stops creating or using new memories, or reset it to permanently delete all memories, including project memories, which cannot be undone. You can also start an incognito chat that is excluded from memory and chat history.
05Is Claude memory available on the free plan?
Yes. Memory from chat history is available to all Claude users, including the free plan as well as Pro, Max, Team, and Enterprise. The same global synthesis and separate Project memory behavior applies across those plans.
The takeaway
Claude runs many brains, not one, and that is on purpose. Global memory is a daily synthesis of your standalone chats, with a direct-note shortcut when you need a fact pinned now; every Project is a sealed room with its own memory that never joins the rest. The wall is the right call for confidential and client-specific work, and the wrong call when you wanted one continuous Claude across everything. Pick your room deliberately. And when you need context that travels past Claude's walls entirely, that is a job for a portable memory layer you control.
