Claude Just Stopped Forgetting Your Project. Mostly.
TL;DR: On 25 August, Anthropic made Claude memory work across chat and Cowork, so what you tell it in one place is there in the other. It also stopped waiting until a conversation ends to remember things. It’s on by default on the free plan. That removes the single most tedious tax on building with AI — re-explaining your project every session — but it comes with one limitation worth knowing and three settings worth opening.
What changed in Claude memory
Two things, per Anthropic’s own post.
Claude memory is now shared. What Claude learns in a chat is available when Cowork runs a task, and what Cowork picks up comes back into chat. Anthropic’s example is being able to lean on months of accumulated context — your “Q3 priorities and the status of your projects” — without setting it all out again.
It remembers as it goes. Claude now “adds topics to memory as you chat, instead of summarizing conversations after they end”. TechCrunch reports that this makes what it knows “more quickly updated and shared”, so moving from working something out to actually doing it doesn’t lose the thread.
It’s on by default on Free, Pro and Max, across web, desktop and mobile — mobile needs the current app version. On Team and Enterprise plans it’s the reverse: admins control it and individuals opt in.
Why this matters more than it sounds
Re-explaining your project is the tax nobody warns beginners about. You spend the first ten minutes of every session describing what you’re building, what you already tried, and why that last approach didn’t work — and you pay it again tomorrow.
We made this the test in our piece on why the scaffolding around a model matters more than the model: “If you have to re-explain your project every session, that’s a weak harness, whatever model is underneath.” This is precisely that problem being addressed, four days later, by a vendor — and on the free plan rather than as a paid upgrade.
For a non-developer, Claude memory carrying across surfaces is worth more than a model upgrade. A slightly cleverer model saves you seconds. Not having to rebuild context saves you the will to keep going.
The catch
Anthropic’s release note is specific about scope: “Memory now works across chat and Cowork in the cloud.”
That wording matters. Cowork launched running locally on your machine in an isolated VM, and the release notes show remote sessions arriving later in beta. So the shared memory is tied to cloud sessions — if you run Cowork locally, don’t assume your chat context follows you there. Worth a check rather than an assumption.
Two smaller things: this is Claude chat and Cowork, not Claude Code, so it isn’t a change to the coding tool itself. And if you’re on a Team plan, you may have it and not know, or not have it and assume you do.
The three settings to open
Shared Claude memory means retention, and retention is worth two minutes of attention. All of this lives in Settings → Memory.
| What to check | Why |
|---|---|
| Topics | Everything Claude has stored appears as a list you can read, edit or delete individually |
| Sensitive topics toggle | Off by default — health, race, ethnicity, religious beliefs, politics and gender identity are excluded unless you turn them on |
| Pause / reset | You can stop memory or clear it at any time; corrections you make apply everywhere from then on |
Some categories are excluded regardless of what you choose: TechCrunch reports Claude never stores government IDs, national insurance or social security numbers, criminal history or immigration status.
That’s a more considered design than we saw elsewhere this month, and worth saying so — we were sceptical about Claude’s watermarking two weeks ago, and this is the same company getting the controls right.
What to do
- Open Settings → Memory and read the Topics list. It takes a minute and it’s the only way to know what Claude actually has.
- Delete anything stale. Old project details will otherwise keep shaping answers long after you’ve moved on.
- Leave sensitive topics off unless you have a reason. Dietary requirements are a fair reason; most things aren’t.
- If you use Cowork locally, verify memory carries. Anthropic’s wording covers cloud sessions specifically.
- Stop opening with a project summary. The habit outlives the need — try assuming it remembers and correct it when it doesn’t.
Who should care (and who shouldn’t)
- Using Claude daily to plan and build: the biggest winners. Less re-briefing, and it’s free.
- On a Team or Enterprise plan: check whether your admin has enabled it before relying on it.
- Using Claude Code specifically: this isn’t your change — Claude Code has its own project memory.
- On Lovable, Bolt, Replit or Horizons: nothing to do, but this is a fair thing to expect from your tool next.
Our take
We’d rather see this than another model release, and we’d say the same about any tool. Memory is unglamorous, hard to market, and worth more to a beginner than a benchmark score — which is exactly why it tends to arrive late.
The honest caveat is that Claude memory is a trade, not a gift. It is genuinely more useful and it is a company retaining more about your work. Anthropic has handled that about as well as it could — visible entries, per-item deletion, sensitive categories off by default, hard exclusions that can’t be toggled on. The controls are real. We’d still open them once rather than take that on trust.
Not sure which AI tool fits what you’re building? Take the 60-second Vibe Coding Tool Finder quiz →
Some links on this page are affiliate links. If you buy through them we may earn a commission at no extra cost to you.
FAQ
Is Claude memory free?
Yes. It’s on by default on the Free, Pro and Max plans, across web, desktop and mobile, though mobile users need the latest app version. Team and Enterprise plans work the other way round — administrators control availability and individual users have to enable it themselves.
Can I see and delete what Claude remembers?
Yes. Everything is listed under Topics in Settings → Memory, where you can read, edit or delete each item individually, and pause or reset memory entirely. Corrections apply everywhere from that point on. Some categories are never stored at all, including government ID numbers, criminal history and immigration status.
Does Claude memory work in Cowork if I run it locally?
Anthropic’s release note says memory works across chat and Cowork “in the cloud”, and Cowork can run either locally in an isolated VM or remotely. So don’t assume shared memory follows a local session — check before relying on it for anything important.
