Claude can now remember past chats — but only when asked

Coming back to work after a vacation and re-explaining an entire project to an AI is easily one of the most tedious parts of using chatbots.

On Monday, Anthropic rolled out a long-awaited memory feature for Claude. The chatbot can now search and reference your past conversations across web, desktop, and mobile, meaning you no longer have to start from scratch every Monday morning.

Talk about yesterday

In a demo video, Anthropic showed a user returning from vacation and simply asking Claude what they were working on before leaving. Claude scanned past conversations, summarized the project context, and asked if the user wanted to pick up where they left off.

Anthropic claims this will keep projects moving seamlessly across sessions, while also respecting context boundaries. The feature keeps separate projects and workspaces neatly divided, so your marketing experiments won't bleed into your backend code discussions.

Right now, it is rolling out to paid users on the Max, Team, and Enterprise tiers. You can turn it on under Profile, Settings, and Search and reference chats. Anthropic noted that other tiers will get access soon.

No shadow dossiers

There is a massive catch that sets this apart from the competition: Claude is not building a secret dossier on who you are.

Unlike OpenAI's persistent memory in ChatGPT, Anthropic spokesperson Ryan Donegan confirmed that Claude will only look at your history when explicitly asked. It will not passively generate an ongoing user profile in the background.

This deliberate constraint arrives at a sensitive moment. Memory in AI has become a hot-button issue recently, especially as ChatGPT faces scrutiny over users relying on it as an ersatz therapist, with some observers pointing to disturbing mental health episodes dubbed ChatGPT psychosis.

The feature drop also lands right in the middle of a brutal arms race. OpenAI just launched GPT-5 last week, while Anthropic is reportedly closing a massive funding round that could value the company at $170 billion. Both are desperate to lock in user loyalty, and memory is the clearest path to retention.

Why it matters

While OpenAI wants ChatGPT to feel like an intuitive friend who remembers your coffee order, Anthropic is treating memory strictly as an on-demand filing cabinet.

For power users and teams juggling messy roadmaps, this is an immediate win. You can finally break gigantic initiatives into separate chats without losing the overarching thread, all while keeping corporate workspaces strictly segregated. If you are already paying for Claude Max or Team, flipping this toggle is a no-brainer.

We think Anthropic's opt-in, query-driven approach is a brilliant hedge against liability. By refusing to build background behavioral profiles, Claude sidesteps the creepy anthropomorphism and parasocial traps that OpenAI is currently taking heat for.

However, it also exposes Claude to a product handicap. An assistant that only remembers when commanded feels undeniably colder and more mechanical than one that naturally anticipates your context. It is safer for the enterprise, but it remains to be seen whether everyday users prefer safety over effortless magic.