Blog
Published: October 28, 2026
Yes, GitHub Copilot has memory. Copilot Memory keeps two kinds on GitHub: repository facts, which every collaborator's Copilot can use in that one repository, and user preferences, which follow you between repositories. A memory expires after 90 days unless Copilot reinforces it. VS Code also has local memory files, which are a separate feature. Every fact below is from GitHub's and Microsoft's own pages, read on 28 October 2026.
What does GitHub Copilot Memory store? #
Copilot writes memories itself, as it works. You can't type one into settings, but you can ask for one (“Remember that API handlers in this repository return errors through formatApiError”), and it decides the kind from who the memory applies to (About Copilot Memory, creating memories):
| Repository facts | User preferences | |
|---|---|---|
| What | Conventions, architecture decisions, build and test commands | How you like to work, what needs your approval, your workflows |
| Who sees them | Anyone with read access who can use Copilot Memory | You, anyone shown a shared session that uses one, and admins on Business and Enterprise |
| Where they apply | Only that repository. Not pooled across an org, not copied to forks | All your repositories |
| Who can create them | Users with write access and memory enabled | Only you, from your own sessions |
| Cited with | The code that supports the fact | Your own words |
Memories with a detected secret are rejected. In the CLI and the GitHub Copilot app, Copilot can ask you to approve a memory first. The cloud agent and code review store without asking.
Is Copilot Memory on by default? #
User preferences are on by default for Copilot Pro, Pro+ and Max, and off on Business and Enterprise until an owner enables the policy. Repository facts are allowed on every paid plan unless a setting restricts them, because an organization's repository access setting defaults to all repositories (default settings). Some surfaces add a local switch:
| Surface | Default |
|---|---|
| Copilot CLI, interactive | On, following your GitHub settings |
copilot -p runs |
Off. Add --enable-memory per run |
| VS Code | Off. Turn on chat.copilot.memory.enabled |
| GitHub Copilot app, cloud agent | Follow your GitHub settings |
| Code review, Agentic Autofix | Follow repository settings, and never use your preferences |
GitHub also lists Copilot in GitHub Mobile, Slack and Microsoft Teams as users of memory, and what one surface learns, another can use. Older pages differ: the VS Code docs say memory is on there by default, and the March and May changelogs call it a public preview, a label the memory docs no longer carry. Check the setting rather than the page.
Where is GitHub Copilot memory stored? #
On GitHub, not in a file. Each repository fact belongs to its repository and is listed in that repository's settings. User preferences are stored in your user settings. On Business and Enterprise plans a preference belongs to the organization or enterprise that licensed you when it was created, so if you move to another license, Copilot stops retrieving it for you. With licenses from several places, Copilot creates no new memories until you pick a default billing entity (managing it for yourself).
Copilot Memory vs VS Code memory files #
The memory files people find on disk are not Copilot Memory. VS Code's memory docs split by agent harness: the Copilot harness uses the GitHub service, and the Local harness has its own memory tool that writes files on your machine.
/memories/for you in every workspace. Its first 200 lines load at the start of every session./memories/repo/for this workspace only, on your machine, not a shared project file./memories/session/for the current conversation, where the Plan agent keepsplan.md.
Open them with Chat: Show Memory Files and wipe them with Chat: Clear All Memory Files. Clearing them deletes nothing on GitHub. A third thing with the same name lives in Visual Studio: its Copilot memories (January 2026) detect a preference as you prompt, ask you to confirm, and save it into %USERPROFILE%/copilot-instructions.md or the repository's .github/copilot-instructions.md.
How long does Copilot Memory keep a memory? #
90 days from when it was created or last upvoted. Retrieving a memory does not extend it. Enterprise and organization owners can set repository facts anywhere from 7 to 180 days, and the lower value wins when both set one. Preferences always use the default (retention). The VS Code docs and the earlier changelogs still say 28 days.
Before using a fact, Copilot checks its citations against the current branch. GitHub calls that “a judgment Copilot makes, rather than a guarantee”, and notes that facts can come from pull requests closed without merging. Copilot upvotes memories it confirms and downvotes wrong ones. A memory hidden from the list is not necessarily deleted.
How do I turn off GitHub Copilot memory? #
- For yourself: profile picture,Copilot settings , Features, Copilot Memory. This turns off your preferences only. It does not delete them or turn off repository facts.
- For a repository: an admin opens Settings, Copilot, Memory and switches off “Allow Copilot to store and retrieve repository facts”. Stored facts stop being used, but stay stored.
- For an organization: Settings, Copilot, Policies, Copilot Memory, with a separate Repository access choice of all, selected or no repositories. Enterprises set it under AI controls (admin docs ).
- In the CLI:
/memory offturns off both kinds on that machine,/memory showchecks it, or set"memory": false. Start a new session after: memories already in context stay there. - In VS Code: clear
chat.copilot.memory.enabled.
How do I view or delete Copilot memories? #
Your preferences are under Copilot settings, Memory: the trash icon deletes one, checkboxes and Delete remove several. Repository facts are under the repository's Settings, Copilot, Memory, for anyone with write access and preferences enabled. Telling Copilot to forget something only downvotes the memory; delete it in settings to remove it. Business and Enterprise owners can export members' preferences as JSONL or delete them in bulk.
Copilot Memory vs custom instructions #
GitHub's own advice: state important rules, such as a security requirement or a review step, in custom instructions, and let memory hold the detail that would bloat the file. “Neither memories nor custom instructions guarantee that Copilot will follow a rule every time.” How to write the file, and how it relates to AGENTS.md, is in copilot-instructions.md vs AGENTS.md.
Copilot Dreaming, a public preview, maintains memory in the background and can open a pull request against .github/copilot-instructions.md, AGENTS.md, CLAUDE.md and skills when it finds them out of date. Its first preview needs GPT-6.1 Sol and is not on Copilot Pro. A “memory bank” is something else: a community pattern of markdown files kept in the repo that the agent reads and updates, as in Cline's Memory Bank.
Does Copilot Memory reach Claude Code, Codex or Cursor? #
No. Every surface GitHub lists is Copilot, and apps built on the Copilot SDK can turn memory on but get no API to list or edit it. A fact Copilot learned about your payments service stays inside Copilot. The one way across is a file: once a Dreaming pull request to AGENTS.md or CLAUDE.md merges, the other agents read it. Each of them keeps its own memory too: Claude Code auto memory, Codex memories and what replaced Cursor Memories. All four side by side are in Claude Code memory vs Codex, Cursor and Copilot.
Memory for a team that uses more than Copilot #
Copilot Memory learns from Copilot's own work: a fact stays in one repository, a preference with one person. What your team decided in a Slack thread, and the rules your Claude Code and Cursor users need, are outside it. To keep one set of rules for every tool by hand, see how to share memory and rules across tools.
Harbor is shared memory for AI coding agents. It learns from where your team decides (Slack, pull request reviews, docs and agent sessions), lets you set what waits for a person, and gives Claude Code, Cursor, Codex and GitHub Copilot the rules that apply to each task.
Questions #
Does GitHub Copilot have memory?
Yes. Copilot Memory stores two kinds on GitHub: repository facts, shared with collaborators in that one repository, and user preferences, which follow you across your repositories. Copilot writes them as it works. You can ask it to remember something, but you cannot type a memory into settings.
Is Copilot Memory on by default?
As of 28 October 2026, user preferences are on by default for Copilot Pro, Pro+ and Max, and off on Business and Enterprise until an owner enables the policy. Repository facts are allowed on all paid plans unless a setting restricts them. VS Code and copilot -p runs each need their own switch turned on.
How do I turn off GitHub Copilot memory?
For yourself, open Copilot settings, Features, Copilot Memory. For a repository, an admin switches off repository facts under Settings, Copilot, Memory. In Copilot CLI run /memory off, and in VS Code clear chat.copilot.memory.enabled. Turning it off does not delete stored memories.
How long does Copilot Memory keep a memory?
GitHub's docs say 90 days from when Copilot created it or last upvoted it; retrieving it does not extend that. Organization and enterprise owners can set repository facts anywhere from 7 to 180 days. Older changelogs and the VS Code docs still say 28 days.
Is Copilot Memory the same as VS Code memory files?
No. VS Code's memory tool in the Local harness writes files on your machine, in /memories/, /memories/repo/ and /memories/session/. Copilot Memory is stored on GitHub. Running Chat: Clear All Memory Files deletes nothing on GitHub.