Anthropic’s “Memory” feature enables Claude to recall past AI conversations and user-provided context, allowing the model to create more tailored and consistent responses when working on long-term or related projects.
How it Works: Claude builds a memory based on your previous AI conversations. It doesn’t remember the entire conversation, but rather creates summaries of what it thinks will be useful in the future and uses these summaries behind the scenes with your prompts.
Project-Specific Memory: The standout feature compared to other models is the ability to create distinct memories for different projects. This acts as a safeguard, keeping context for a specific client, team, or initiative completely separate from others.
User Control: You have direct control over what Claude remembers. You can view and edit the memory summary in your settings and explicitly instruct Claude to remember or forget specific pieces of information. Memory is off by default.
Incognito Chat: Claude offers an “Incognito” chat mode for conversations that you do not want saved to its memory or your conversation history. This is ideal for one-off queries or sensitive discussions.
If you’ve read my article on context modules, you understand the value of concise, reusable descriptions for helping you create content that is consistent in style, format, and language, but can be personalized for different audiences and objectives. Claude is essentially creating context modules for you, but with considerably less visibility and control over the specifics.
Whenever you add personal or organization-specific context to a prompt, you risk inadvertently exposing proprietary or private data. Auto-generated context memory can increase the risk because you might not consistently check what’s in AI memory or forget to use the incognito setting when appropriate.
OpenAI (GPT-4 and GPT-5) and Google (Gemini Pro) have similar memory features, but their features work at the user or team level rather than being isolated to pre-defined projects. Also, their memory features are on by default, whereas you have to explicitly turn on the memory feature in Claude.
