Key takeaways
What happened
Today, Anthropic announced the merger of the memory systems for Claude chat and Claude Cowork. "Starting today, the memory you use in chat is the same as in Claude Cowork. Now, wherever you work with Claude, it starts from what it already knows about you," said Anthropic. Memory is now two-way. What you do in Claude chat is remembered by Claude Cowork.
" That said, you might want Claude to know about topics some consider sensitive. Anthropic's example was that you might want Claude to remember you have a gluten allergy so it takes that into account when suggesting recipes. If you've granted Claude permission to save sensitive topics, Claude will display a notice. On one hand, that's helpful, so you know information is being recorded.
But if you're struggling with something and just want some help, I'm not sure getting a warning about it every time you want to discuss it will be conducive to increasing your internal peace. Anthropic lists a number of topics that it will not store, even if you allow sensitive topics to be stored.
These include sensitive identification numbers, criminal history, immigration status, or anything that violates Anthropic's acceptable use policy. These memory features are available in all tiers of chat and Cowork. " However, saving sensitive topics is off by default. Enterprise and Teams tier users are a bit different. " Controls for the memory features are available directly from the Settings menu in the new Memory section.
Why it matters
What you do in Cowork is remembered in chat. It seems that what happens in Claude Code still stays in Claude Code. The company made no mention of Claude Code whatsoever in the announcement. " One of the side effects of this rapid pace of innovation is that experiments have turned into projects and then into products, often without the benefit of an end-to-end product plan.
As a result, we've been seeing capabilities mirrored across products but not necessarily coordinated between them. Features appear in one implementation but mysteriously vanish in the next (ChatGPT desktop, I'm looking at you). We're also starting to see the consolidation of features.
This can be best seen in the trend, on the part of both OpenAI and Anthropic, to turn their individual desktop app offerings into "super apps," apps that contain the features of more than one previously individual desktop app. Now, we're seeing another consolidation. Up until now, the memory of chat sessions you had with Claude as a chatbot was locked into the chatbot.
The memory of sessions you've had with Claude Cowork was locked into Cowork. " This implies that Claude's existing memory of your conversations will be available to Cowork, not just conversations that you start going forward. The company gives an example that supports this idea of preexisting knowledge: "Ask Cowork to draft an update for your manager, and it already knows who that is and how she likes updates written.
Brainstorm the agenda in chat for a conference you're organizing; when Cowork builds the budget and logistics doc, it knows the headcount, the city, and the speakers. " One interesting operational change is how Claude adds topics to memory. They are now added to memory as you chat. Previously, Claude used to summarize conversations after they ended. This is dynamic.
You don't have to tell Claude to "remember this" for it to remember. This update includes a number of safety and quality-of-life improvements to memory as well. You can pause or reset memory at any time. That means nobody needs to know about your unhealthy fascination with British cozy mysteries or 1980s hair bands. In all seriousness, Claude also has fairly deep controls over sensitive topics.
What to watch
To control whether Claude keeps track of your chat memories, toggle "Generate memory from chats" on or off. To control sensitive memory retention, toggle "Include sensitive topics in memory" on or off. You can also go back and edit your saved memories. Claude stores a list of files under Topics in the Memory settings area. You can read, edit, or delete them. They apply globally.
The example Anthropic used was that if you correct your company's "old name," then every conversation afterward will know the new name. I have to tell you, I am seriously conflicted about this. But then again, I'm seriously conflicted about the whole AI thing.