Let us begin with a plain definition, because the word “memory” can mean several different things.

Memory in Claude is a feature that lets Claude retain useful context, such as your preferences, ongoing projects and working style, across separate conversations. Instead of starting every chat knowing nothing about you, Claude can begin with a short, editable picture of the things that help it assist you better.

It helps to know what memory is not. It is not Claude secretly recording every word you say and replaying it. It is not the same as the “context window”, which is the working space inside a single conversation. And it is not a human-style memory with feelings and personal history. It is better understood as a set of notes that Claude can read at the start of a chat.

Without memory, every conversation begins with the same tedious introduction.

Claude, like other language models, does not carry anything from one conversation to the next by default. When a chat ends, the working space is cleared. This is why, before memory, people often had to paste in the same background again and again: their job, their audience, their favourite writing style, the project they were working on.

Memory reduces that repetition. If you often ask for help with the same kind of work, Claude can start from a more informed position. A student who always wants simple explanations, a manager who writes weekly updates, or a developer with a preferred coding style can all benefit from not restating those details every time.

flowchart LR
  A["Without memory"] --> B["Start each chat from a blank page"]
  B --> C["You repeat your background every time"]
  D["With memory"] --> E["Claude starts with useful saved context"]
  E --> F["You can get to the real question sooner"]
        

Memory is designed mainly for useful, work-relevant context, not for every detail of your life.

According to Claude’s help centre, memory focuses on things that help Claude assist you, such as your communication preferences and working style, technical preferences and coding style, and details about your projects and ongoing work. The idea is professional and practical context, such as how you like answers presented and what you are working on.

Memory is built from your conversations, and how it is kept can differ between plans and versions of the feature. In the current design, Claude is described as building individual memory entries as you chat, grouping them into categories, and letting you read, edit or delete them. Earlier versions worked differently, with a summary that updated on a roughly daily cycle. Because the design has changed, check your own Memory page to see how it looks for you.

You do not need the engineering details, but a simple picture helps you use memory wisely.

Claude’s model itself does not change when you chat. Its core learning is fixed. Memory is a separate layer added around the model. As you converse, useful details can be saved as notes. When you start a new conversation, relevant notes are placed in front of Claude, so it can use them while writing its answer. In other words, memory works by supplying Claude with saved text at the start of a chat, not by rewiring Claude’s brain.

flowchart TD
  A["Conversation 1"] --> B["Useful details saved as memory"]
  B --> C["Memory store you can view and edit"]
  C --> D["Conversation 2 starts with relevant memory in view"]
  D --> E["Answer shaped by your saved context"]
        

These three features are often confused, so let us separate them clearly.

Memory gives Claude context about you, carried across conversations. Projects give Claude context about a specific body of work, such as a set of files and instructions that stay available to chats inside that project. Chat search, sometimes described as “search and reference past chats”, lets Claude look back through earlier conversations when you ask, to retrieve specific details. Per the help centre, searching past chats is available on paid plans and is limited to the boundaries described, such as chats outside projects, or within a single project.

flowchart TD
  A["Your Claude account"] --> B["Memory: about you"]
  A --> C["Projects: about one body of work"]
  A --> D["Chat search: find specific past chats"]
  A --> E["Incognito: nothing saved"]
        

Projects add an important twist, because they can keep their own separate memory.

A project is a saved workspace for a body of work. According to Claude’s help centre, each project can have its own separate memory space and its own project summary, so that context in one project stays focused and separate from other projects and from your non-project chats. This helps keep, for example, work for one client from mixing with a personal hobby or another client’s work.

There is also a project-level setting that controls how a project’s memory relates to your wider account memory. Depending on that setting, a project’s memory can be kept entirely separate, or connected so that the two can draw on each other. Because this is a per-project choice found on the project’s own page on the web, look there if you want to change it, and consult the help centre for the current options.

Availability and defaults depend on your plan and on when the feature reached it.

Memory rolled out in stages. Team and Enterprise plans received it first in September 2025, Pro and Max followed in October 2025, and it reached free users in early 2026, according to coverage and help pages at the time. The help centre says that on Free, Pro and Max plans memory applies on the web, Claude Desktop and the mobile apps and is on by default, while on Team and Enterprise plans it is off by default and an owner can turn it on. It is not currently available in Claude’s separate agent-style work environment known as Cowork.

Anthropic has also redesigned memory more than once. Coverage in 2026 describes a redesign in July, a dedicated Memory page in August, and the closing of an export window for the older version in September. Since the details evolve, treat this section as a snapshot.

The most important feature of memory may be your ability to manage it.

Anthropic describes memory as fully optional, with granular controls. You can generally open your Memory page to see what Claude has stored, edit entries or the summary directly, delete things you do not want kept, and tell Claude explicitly what to remember. You can also turn memory off, pause it, or reset it. The user can see memory tool activity in the chat when Claude reads from it, depending on the product.

Sometimes you want Claude’s help without leaving any trace in memory.

An incognito chat is a temporary conversation that is not saved to your chat history and does not contribute to memory. According to the help centre, incognito chats are available on all plans, and you start one by selecting the ghost icon in a new chat outside a project. Claude does not draw on incognito chats when searching your previous conversations.

There is an important caveat. Incognito does not mean that nothing exists anywhere. For example, the help centre notes that on organisational plans, incognito chats can still be included in organisational data exports and follow the organisation’s data retention policies. Anthropic’s materials also describe a short default retention period for incognito chats. So incognito is best understood as “not saved to my history or memory”, not as “completely invisible to everyone”.

flowchart TD
  A["Start a conversation"] --> B{"Incognito?"}
  B -- "No" --> C["Saved to history and may contribute to memory"]
  B -- "Yes" --> D["Not in your history and not added to memory"]
  D --> E["Organisation or retention rules may still apply"]
        

Memory can often be moved between Claude and other AI assistants, with some caution.

Anthropic offers a way to bring memory details over from another AI assistant, or to export your memory from Claude for backup or migration. In outline, you ask the other assistant to list what it has stored about you, then paste that text into Claude’s memory import tool, which adds the entries. The company describes this tool as experimental and still in active development, so imported details may not always be handled perfectly. The tool’s exact location can change, so use the instructions in the current help centre article.

A few simple habits help memory work for you instead of against you.

Memory is useful, but it comes with limits and responsibilities on both sides.

Privacy deserves thought. What you tell Claude in a conversation may become part of memory, so consider what you are comfortable having saved. If you use Claude through an organisation, its administrators may have controls and retention policies that apply to memory-related data. Read the privacy information for your plan, and use incognito chats for conversations that should not be saved.

These errors appear often, and each has an easy fix.

Here is the whole picture in one place.

Yes. Claude can retain useful context, such as preferences and ongoing work, across conversations. The feature is optional, and what it looks like and where it is available has changed over time, so check the current help centre article.

No. It is a set of notes or a summary of useful context, not a full transcript. Details can be missing or simplified.

Yes. You can open your Memory page in Settings to view, edit or delete what is saved, and you can also tell Claude what to remember or forget.

You can turn memory off in your Settings, or use an incognito chat for individual conversations. Claude cannot switch the setting off by itself, so use your Settings for that.

A temporary conversation, started with the ghost icon, that is not saved to your chat history or memory. Organisational plans may still apply export and retention rules.

Each project can have its own separate memory space and summary, which keeps its context apart from other projects and standalone chats. A per-project setting on the web controls how a project’s memory relates to your wider memory.

Memory is a saved picture of your preferences and context applied in the background. Chat search lets Claude look back through earlier conversations when you ask. Searching past chats is available on paid plans.

Per the help centre at the time of writing, it is on by default for Free, Pro and Max plans, and off by default on Team and Enterprise plans until an owner turns it on. Availability has changed over time, so confirm for your plan.

Anthropic offers an import and export path, described as experimental. Review any text carefully before importing, and check the results on your Memory page afterwards.

The article “Understanding Claude’s memory” at support.claude.com is the best source, along with the Memory section of your own Settings. Developers can find separate guidance on memory tools in the documentation at docs.claude.com.

Claude’s memory lets useful context travel from one conversation to the next, so you do not have to introduce yourself every time. It works by saving helpful notes, such as your preferences, working style and ongoing projects, and making the relevant ones available at the start of later chats. It is different from the context window, which exists only inside one conversation, from projects, which hold the context of a particular body of work and can keep their own memory, and from chat search, which retrieves specific past conversations on request. You stay in control by viewing and editing your memory, telling Claude what to remember or forget, turning memory off in Settings, and using incognito chats when you want nothing saved. Memory is a convenience, not a perfect record, so restate key facts for important tasks, review your notes regularly, keep secrets out of it and check the current help centre article, because this feature has changed often.