Copilot Memory
Persistent user preferences and repository knowledge that Copilot carries across sessions
title: "Copilot Memory" description: "Persistent user preferences and repository knowledge that Copilot carries across sessions" section: "Copilot" readTime: "4 min" badge: "NEW"
Copilot Memory
Copilot Memory lets GitHub Copilot remember things about you and your repositories across sessions. Instead of re-explaining your preferences every time, Copilot stores them and applies them automatically.
On individual plans Copilot Memory is on by default. On organization- and enterprise-managed plans an administrator must first enable the policy, after which users can opt out. Business and Enterprise administrators can export or delete user-level preferences.
Two Memory Scopes
User-level preferences
Preferences tied to you, stated or implied by your working style. They apply only to your interactions, across repositories.
Examples:
- Your preferred commit message style
- How you like pull requests structured
- Preferred test framework or naming conventions
- Whether you prefer verbose or terse explanations
Repository-level facts
Facts tied to a specific repository, such as coding conventions, architecture decisions, and build commands. Any user with Copilot Memory access to that repository benefits, but the facts apply only within it. Repository owners can review and delete them.
How It Works
Copilot captures preferences from what you state directly ("Always use conventional commits format") and from patterns it notices in how you work.
Memories are validated before use:
- Repository facts are linked to supporting code, and Copilot checks those citations against the current branch
- User preferences are linked to citations (which may include your own words), and Copilot judges whether each still applies
- Facts drawn from pull requests closed without merging are used only if the current codebase still supports them
A stored fact or preference that goes unused is deleted after 28 days. Successful validation and use can reset the timer.
Where It Is Used
- Copilot cloud agent
- Copilot code review (repository-level facts only)
- Copilot CLI (repository-level facts plus the initiating user's preferences)
- Agentic autofix
Facts captured by one surface can be used by the others.
Manage Copilot Memory
Memory is turned on per user, not per repository. Manage it from your personal Copilot settings on GitHub.com (see Managing Copilot Memory in the docs). If you hold multiple Copilot licenses, choose a default billing entity so user-level preferences can be generated.
Copilot Memory vs Claude Code Auto Memory
Both tools have persistent memory systems, but they work differently:
| Copilot Memory | Claude Code Auto Memory | |
|---|---|---|
| Scopes | User + repository | Repository (per git root) |
| Storage | GitHub cloud | Local machine |
| Syncs across devices | Yes | No |
| User editable | Via GitHub settings and repo owner review | Plain markdown files |
| Expiry | Unused facts deleted after 28 days | Persist until you remove them |
| Used by | Cloud agent, code review, CLI, autofix | Claude Code only |
Copilot Memory travels with you across devices and Copilot experiences. Claude Code auto memory stays on your machine but is fully transparent: plain markdown you can read and edit directly.
Related
- Customization — Custom instructions and agents
- Agents — Agent types and sessions
- Cloud Agent — A surface that uses repository memory