GitHub Copilot Memory, dreaming and code search expand agent context
GitHub introduced GitHub intelligence, a context layer for Copilot agents. Copilot Memory is now generally available on paid plans, dreaming is in public preview, and Copilot can search code across repositories without a local clone. Agents can now keep your preferences and team conventions between sessions and reach related projects.
Key points
- Copilot Memory is generally available on paid Copilot plans and shared across Copilot surfaces
- Dreaming is in public preview and opt-in, with personal and repository settings configured separately
- Changes to checked-in skills and instructions are proposed as pull requests, not written directly
- Organizations enable dreaming through Copilot policies; enterprise admins can delegate that choice
- Dream runs consume AI credits, so check the costs before enabling it broadly
- Code search covers related repositories without a local clone, using exact, regex or semantic queries
GitHub argues that coding agents should know more than the current codebase; they should also know how a team works. It announced GitHub intelligence, a context layer built on GitHub's data graph, with three parts. Memory carries preferences and repository conventions between sessions and clients. Dreaming reviews past work to keep that context current. Code search lets Copilot find code, configuration and docs across related repositories without a local clone.
Memory is now generally available on paid Copilot plans and works across the Copilot app, CLI, VS Code, cloud agent, code review, Slack, Teams and GitHub Mobile. Dreaming is in public preview and is opt-in: it can update personal and repository memories, while changes to checked-in skills and instructions arrive as pull requests for review. Dream runs consume AI credits. Code search supports exact symbols, regular expressions and semantic queries, and works with the /research command to trace implementations across services.
Real-world use
- Tracing retry logic across a payments service, a shared library and background workers before changing it
- Keeping test commands and team preferences in memory instead of repeating them in every session
- Finding which projects in the organization still call an old API and planning the migration
Why it matters
Your agents can reuse team conventions and reach related repositories without repeated setup, but you should review memory access, dreaming costs and the pull request flow for skill and instruction changes before rolling it out.