GitHub Copilot Memory goes GA as Dreaming enters preview
GitHub announced general availability for Copilot Memory on paid plans on October 28, alongside a public preview of Copilot Dreaming. They are related but different: Memory carries preferences and repository facts between sessions, while Dreaming maintains context in the background. Neither means the underlying model is being retrained for your repository.
The announcement groups these capabilities with code search under GitHub intelligence. The useful question for a team is narrower than the umbrella name: what can be remembered, who controls it and which changes happen without a pull request? Availability differs by capability.
Memory has more than one switch
The Memory guide distinguishes repository facts from user preferences. Facts belong to one repository and are not pooled across an organization. Preferences can follow a user across repositories, subject to ownership and policy. On Business and Enterprise plans, the billing organization or enterprise owns preferences and can view, export or delete them.
Personal preferences default on for Pro, Pro+ and Max. On Business and Enterprise, they start disabled until an administrator enables the policy. Repository facts have separate access controls. Turning off personal preferences does not turn off repository facts. VS Code also has a local setting that defaults off.
Memory availability in the Copilot app still depends on the account. Its project-memory setting is a separate feature, not a control for these shared facts and preferences. Stored facts are not guaranteed rules. GitHub recommends custom instructions for guidance you want provided consistently.
Dreaming changes are not all pull requests
The Dreaming preview documentation requires an eligible paid plan and access to GPT-6.1 Sol. It excludes Copilot Pro and Enterprise Cloud with data residency. Dreaming is opt-in. Required Memory settings, session syncing and organization policies depend on the capability.
| Output | How it changes |
|---|---|
| Personal or repository memories | Added, updated or deleted automatically |
| Checked-in instructions and skills | Proposed in a pull request |
| Internal code-review guidance | Updated without a pull request |
Every Dreaming run consumes AI credits, including runs that make no changes or process inactive repositories. Turning it off stops future runs but does not undo earlier changes or necessarily stop a run already underway. Disabling session syncing does not delete previously synced sessions.
Test the distinction before expanding access
An illustrative pilot could use a disposable repository whose documented test command has intentionally fallen behind its actual configuration. Ask whether the generated suggestion identifies the mismatch, cites the right files and proposes a correction a maintainer would accept. This is a proposed test, not a claim that we ran it or that Dreaming will always detect that problem.
Keep a short review log: what changed, whether the change required a pull request, whether it was correct and what the run cost. A useful suggestion and a safely governed deployment are different outcomes. Review both before adding more repositories.
ByteForward reviewed the release and documentation but has not enabled or tested Memory or Dreaming. We have not adopted GitHub’s preview-session improvement percentages as evidence of results in a new team’s workflow.
AI-generated editorial illustration by ByteForward. Current Copilot logo reference from the GitHub Brand Toolkit. GitHub marks belong to GitHub, Inc. No affiliation or endorsement is implied.
