- Copilot Memory stores memories about the user and context to personalize responses, securely storing them in Exchange and SharePoint Embedded.
- It is possible to manually add and manage memories and custom instructions in Microsoft 365 Copilot, with controls to activate, deactivate, or delete the memory.
- Administrators have options in Purview, eDiscovery, and Microsoft Graph to search, export, or delete data from memory, although with limits on retention and auditing.
- GitHub Copilot incorporates its own memory per repository, configurable at the company, organization and user level, and with tools to review and delete code memories.
Copilot's memory has become a key component for maximizing the AI capabilities of Microsoft 365 and GitHub. Beyond simply answering your questions in real time, Copilot can remember certain information about you, your work, and your repositories to provide more relevant answers, save you from repetitive steps, and better adapt to the context.
If you're wondering how to manually save information in Copilot Memory, what data is saved, where it's stored, and how to manage (or delete) it at both the user and organizational levels, here's a complete guide based on official documentation and practical experience with the various Copilot variants.
What exactly is Copilot memory and what is it used for?
When you interact with Microsoft 365 Copilot Chat, the AI learns details about your role, your usual tasks, and your preferences . Some of this information becomes "memories" that Copilot can reuse in future conversations so it doesn't always have to start from scratch.
Within Microsoft 365 Copilot, these memories are divided into several types: saved memories, details inferred from chat history, and custom instructions . Together, they make up the "Copilot Memory," which is the additional context Copilot uses to better tailor its responses.
In parallel, GitHub offers another feature called Copilot Code Memory , designed to help the assistant better understand your codebase and repository. This memory is used by the Copilot cloud agent, Copilot code review, and the Copilot CLI to work more efficiently on a specific repository.
It's important to distinguish between what Copilot remembers about you as a Microsoft 365 user (your role, language preferences, the types of tasks you frequently perform) and what it remembers about a code repository on GitHub (rules, patterns, design decisions, etc.). Both use the concept of "memory," but they are managed and stored differently.
How to manually save information to Copilot Memory in Microsoft 365
In the Microsoft 365 environment, memory isn't just automatic. You can tell Copilot what you want it to remember so it can take it into account in future chat sessions.
To manually save information to Copilot's memory from Microsoft 365 Copilot Chat, you can follow this approach:
- Log in to Copilot Chat with your Microsoft 365 work or school account.
- In the writing box, Write explicitly what you want me to rememberFor example: “Remember that I work in the digital marketing department and that my primary language is Spanish from Spain.”
- When Copilot detects that this is relevant data, It will show you a notification asking if you want to save that information. as a memento for the future.
- If you accept, Copilot will add that information to your memories saved in Copilot Memory, and you'll see a message like "Memory updated" confirming that it has been saved.
You can also use phrases like “remember that…” or “add to your memory that…” to directly tell it that you want that information stored. Copilot will automatically manage these reminders, merging related items, updating outdated ones, or deleting any you request.
Keep in mind that the "Save memories" option is enabled by default in Microsoft 365 Copilot , although administrators can control memory usage through the "enhanced personalization" settings.
Where is Copilot's memory stored and how is it protected?
In Microsoft 365, Copilot memories (saved memories, details inferred from chat history, and custom instructions) are stored in your own Exchange mailbox , inside a hidden folder.
This means that Copilot's memory inherits the same security and compliance policies as any other data in your mailbox: encryption at rest, possible use of Customer Lockbox, access controls, etc. From a governance perspective, it's treated as email or mailbox items.
In the case of Copilot Pages and Copilot Notes (page/notebook-type content that you create within the Microsoft 365 ecosystem), storage is different: they are saved in SharePoint Embedded in a user-owned container . This container also hosts the content of Loop's "My Workspace" and is unique per user.
The container name may vary depending on which application the user opens first: “Pages” if they start with the Microsoft 365 Copilot app , or “My Workspace” (localized to Loop's language) if they open the Loop app first. For administrative purposes, it is managed using SharePoint Embedded tools and is identified as belonging to the Loop application.
This container has a maximum limit of 25 TB , participates in the organization's SharePoint storage quota, and follows the same lifecycle as OneDrive: it is deleted when the user account is deleted, goes through a temporary deletion period, and is then permanently purged after 93 days if it is not recovered.
How enhanced customization and memory controls work
The memory and personalization functionality in Microsoft 365 Copilot is governed by a feature called “enhanced personalization .” This control determines whether Copilot can use memories and personalized data to tailor its responses.
By default, enhanced personalization is enabled for end users . However, the tenant administrator can disable it globally or for specific groups through relevant policies, including the enhancedPersonalizationSetting resource in Microsoft Graph, which allows for automated configuration using scripts.
When enhanced personalization is disabled by the administrator, user controls for custom instructions, saved memories, and chat history appear disabled in the Settings > Personalization section, and the user cannot re-enable them.
However, if the organization keeps enhanced personalization enabled, each person can adjust their own controls for:
- Custom instructions: What information about your profile and work style do you want Copilot to always keep in mind?
- Memories kept: what information have you explicitly asked me to remember.
- Chat history: what use can Copilot make of past conversations to infer details.
An important point: the administrator doesn't need to do anything to "activate" Copilot's memory ; it's enabled by default unless explicitly disabled. However, they can disable it at the tenant level or for specific users if the organization requires it.
How long does Copilot retain its memory and what are the limits?
Copilot's memory retention depends on the type of data we're talking about: custom instructions, saved memories, or chat history.
In the case of custom instructions , if the administrator disables enhanced personalization or if the user disables this setting, Copilot simply stops applying them to responses, but does not delete them from the Exchange mailbox. They remain stored there, although inactive from the model's perspective.
For saved memories , the rule is very clear: they are retained until the user explicitly deletes them from Settings > Personalization > Saved Memories. Deleting a chat does not delete the memories that originated from that chat. If enhanced personalization or the Saved Memories control is disabled, Copilot will not use them in context, but it also won't delete them automatically.
Regarding chat history , the inferred details are dynamic. Copilot may update, compact, or discard older information as it learns what is most useful to retain. If you delete all chats in which certain information was mentioned, all traces of that information will be removed from the inferred details within seven days.
If enhanced personalization is disabled or Chat History control is turned off at the user level, then all chat history details are deleted after 30 days . In that scenario, Copilot stops building memory from old conversations.
A key compliance point: retention policies and labels configured in Microsoft Purview (for example, “delete after three months” for Copilot Chat interactions) do not apply to Copilot memory . There are no specific administrative controls to enforce retention rules on these memories; their lifecycle follows the rules described above.
Managing and manually deleting memories in Microsoft 365 Copilot
As a user, you always retain direct control over what Copilot remembers about you . Once memories are saved, you can review and delete them directly from the Copilot Chat interface.
The standard workflow for managing and deleting memories in Microsoft 365 Copilot is as follows:
- Accede to Copilot Chat with your work or school account.
- In the top right corner of the app, open the menu “… Copilot chats and much more”.
- In that menu, enter the Copilot settings.
- Within Settings, go to the section Personalization. and look for the icon or section of Memories kept.
- Press on “Managing saved memories” to see the complete list.
On this panel you can take two types of actions: deactivate Copilot memory using the Saved Memories switch or delete specific memories (one by one) and even delete all memories at once using the "Delete all memories" button.
It's important to understand that disabling memory doesn't automatically delete existing memories . It only prevents Copilot from using them in future responses. If you want them to disappear, you'll need to delete them individually or use the "Delete all memories" option in the dashboard.
If your intention is to stop Copilot from learning from chat history , in addition to deleting memories, it's advisable to review the history-based customization settings and, if necessary, disable that part as well.
Capabilities and limitations for administrators: search, deletion, and auditing
Tenant managers have several tools at their disposal to respond to stakeholder requests related to Copilot memory and to exercise control over this data when necessary.
First, you can use Microsoft Purview eDiscovery and content search to locate, export, and delete user memory data. In these cases, for Copilot memory, you should search within the <IPM.Contact> element class , which is where saved memories and inferred chat history memories are stored.
Purview eDiscovery and the Microsoft Graph Explorer can be used to :
- Locate stored memories and inferred memories associated with a user.
- Export this data to handle access or portability requests.
- Delete specific memories when there is a legal basis or regulatory requirement for it.
Custom instructions cannot currently be detected through eDiscovery or content search. Only the user can manually export them from Settings > Personalization if they need a copy or want to review them.
Another significant limitation is that actions involving memory and customization do not generate audit entries in Purview. In other words, there is no specific audit log for "save a memory" or "delete a memory" within the standard compliance logs.
Finally, administrators cannot limit the exact type of information a user adds to their memory . They can globally disable memory or enhanced personalization, but there is no granular setting that says, “This is allowed to be remembered, but not that.”
Copilot's memory on GitHub: repositories, companies, and organizations
In the world of development, GitHub incorporates a feature called Copilot Memory aimed at improving the performance of the cloud agent, code review, and the Copilot CLI within a specific repository.
The idea is that Copilot can remember relevant details about the codebase : adopted conventions, key components, architectural decisions, commonly used patterns, etc. That way, when you return to work on the same repository, Copilot already understands the context better and you don't have to explain it repeatedly.
For users with an individual Copilot subscription (Copilot Pro or Copilot Pro+), Copilot's code memory is enabled by default . These users can go to their personal Copilot settings on GitHub and choose whether to keep it enabled or disable it.
In corporate environments, with subscriptions managed by companies or organizations, Copilot memory is initially disabled and must be explicitly enabled in the company or organization settings before users can take advantage of it in their repositories.
How to activate Copilot Memory on GitHub (company, organization, and user)
For a company that centralizes multiple organizations on GitHub, enterprise owners can define a Copilot memory enable policy for the entire company or delegate the decision to the owners of each organization.
The general steps for the company level are:
- Go to the section of the company on GitHub (for example, from the "Companies" page on GitHub.com).
- At the top, enter the AI controls.
- In the sidebar, open the section Copilot.
- Within the “Features” section, look for the settings for Copilot's Memoir and choose a policy from the drop-down list:
- Let the organizations decide (each organization will choose whether to activate it or not);
- Enabled globally (all organizations within the company with Copilot licenses will have active memory);
- Disabled everywhere (no memory is allowed and organizations will not be able to change it).
At the organizational level, owners can enable or disable Copilot Memory for all members with a Copilot license. If the organization is under a company, company-level policy may restrict this decision (for example, if company-level policy has set it to disabled everywhere, the organization will not be able to enable it).
The typical workflow for an organization is:
- In the top right corner of GitHub, click on the profile picture and then in “Your organizations”.
- Select the desired organization.
- Below the organization's name, enter Settings (If the tab does not appear, open the drop-down menu and choose “Settings”).
- In the sidebar, under “Code, planning and automation”, go to Copilot > Directives.
- In “Features”, scroll down to the settings for Copilot's Memoir and select “Enabled”.
For an individual user with a Copilot Pro or Pro+ plan, simply:
- Go to any GitHub page, click on the profile picture and then in “Copilot settings”.
- In the “Features” section, locate the option to Copilot's Memoir.
- Choose from the drop-down menu whether you want "Enabled" or "Disabled".
From that point on, as long as Copilot's memory is enabled, it will be used in any repository where you use the cloud agent, code review, or CLI . In this way, Copilot will collect and use memories from that codebase.
Reviewing and deleting Copilot memories in a GitHub repository
As the owner of a repository where Copilot's memory is active, you have the option to review the memories Copilot has stored about that project . This is useful for ensuring that what the AI remembers is correct, relevant, and does not contain inappropriate information.
To view the memories of a repository on GitHub:
- Navigate to the repository homepage on GitHub.
- Below the repo name, go to the tab Settings (or in the "Settings" drop-down menu if it does not appear as a tab).
- In the “Code and Automation” section of the sidebar, open Copilot Memory.
- You will see a chronological list of memories, with the most recent ones at the top.
If you detect a memory module that shouldn't be used (because it's incorrect, outdated, or problematic), you can remove it . To do this:
- In the list of memories, click on the trash can icon located to the right of the memory you want to erase.
- If you need to delete several at once, select the checkboxes Select the memories you want to delete and then click "Delete".
Even so, it's worth noting that Copilot validates memories before using them . It only utilizes a memory if the code that originated it still exists in the repository. If that code snippet has disappeared or changed drastically, the memory may no longer be applicable even if you don't manually delete it.
This combination of manual review by the owner and automatic verification by the tool helps to keep Copilot's memory useful and reliable without accumulating outdated information indefinitely.
Saving messages and prompts in the Copilot Query Gallery
In addition to the "classic" user or code memory, Microsoft 365 Copilot includes a very practical feature: the ability to save frequently used prompts . It's not exactly the same as personal memory, but it serves a similar purpose by "reminding" you of common tasks.
The process is simple. If you create a message that you find useful, such as "Review and summarize today's emails and provide a list of action items," you can save it directly from the chat to have it readily available.
Once Copilot has responded, simply hover your mouse over the message to see the bookmark icon. Clicking "Save Message" will add that prompt to the "Your Prompts" section within the Copilot Query Gallery.
At this point, you'll have the option to edit the message title to make it more descriptive. For example, you could rename it "Daily Email and To-Do Summary" to find it more quickly among your reminders.
The Prompt Gallery also houses a collection of universal prompts selected by Microsoft . These prompts are common to all users in the organization and act as general templates. You can use them as is, save them to your prompts using the bookmark feature, or adapt them to your needs before adding them to your saved messages.
To view all your saved prompts , simply open the Copilot Query Gallery (making sure you're logged in), go to the "Your Messages" tab, and browse the list. From there, you can view, edit, or run any message you've previously saved. It's a very straightforward way to standardize daily tasks, such as creating reports with Copilot in Excel without having to rewrite the same instruction every time.
At this point, you'll have the option to edit the message title to make it more descriptive. For example, you could rename it "Daily Email and To-Do Summary" to find it more quickly among your reminders.
Taken together, Copilot's memory, customization, Exchange and SharePoint Embedded storage, admin controls, and GitHub-specific options create a robust ecosystem. Understanding how all this information is stored and managed allows you to better leverage AI while maintaining privacy, compliance, and control over your own data.
Passionate writer about the world of bytes and technology in general. I love sharing my knowledge through writing, and that's what I'll do on this blog, show you all the most interesting things about gadgets, software, hardware, tech trends, and more. My goal is to help you navigate the digital world in a simple and entertaining way.

