Copilot Memory vs Dreaming: What Changes Automatically?

What can GitHub Copilot remember—and what can Dreaming change without a pull request? In 96 seconds: saved context, source checks, background maintenance and the controls worth checking before you opt in. The key distinction: Dreaming can update memories automatically. Changes to checked-in instructions and skills are proposed through a PR. Maintaining context is not training the underlying model. 00:00 Can an old memory mislead an agent? 00:15 What Copilot Memory carries forward 00:29 Check the source, not just the saved fact 00:40 What Dreaming maintains 00:54 Automatic memory changes vs reviewed files 01:09 Availability and AI credits 01:22 Read. Change. Cost. Sources — checked 7 November 2026: https://docs.github.com/en/copilot/concepts/prompting-and-context/copilot-memory https://docs.github.com/en/copilot/concepts/prompting-and-context/copilot-dreaming https://github.blog/changelog/2026-10-28-github-intelligence-puts-your-code-context-and-history-to-work/ Dreaming is optional and in preview; availability depends on your plan. A dream run consumes AI credits even when it makes no changes. Features and terms may change after the source-check date. Original animated diagrams and sound design, with synthetic English narration. The test-command example is illustrative, not a recording of a real Copilot failure. This is an independent explainer, not a GitHub product demo or sponsorship. #GitHubCopilot #AIContext #AgenticAI