GitHub Copilot CLI now indexes every C++ file in a repository, not only the ones open in the editor

A C++ engineer with a repository of a million lines has, until this release, relied on Copilot's answers being informed only by the files open in the editor at the time. GitHub announced on 22 September 2026 that Copilot CLI now indexes every C++ file in a project through the Microsoft C++ Language Server, so a request for a definition or reference no longer stalls on the wait for the surrounding files to load.

What the feature actually does

GitHub describes whole codebase indexing as a persistent index of symbols across a C++ project, "including files that aren't currently open". The company writes that the Microsoft C++ Language Server "uses your project's compilation information to resolve types, symbols, includes, and relationships between files", and says the effect for a developer is that they "spend less time waiting for definitions, references, implementations, and symbol search results".

The feature is on by default. A user can watch the index build through /lsp logs, and can turn indexing off through documentation on the Microsoft C++ Language Server repository, with a session restart to pick up the change. GitHub does not publish a benchmark number in the changelog, so any speed claim about the feature would be a guess from outside the announcement.

What the cost is

The trade GitHub asks for is time and memory during the first build. The changelog states that "building the index for the first time can take additional time and temporarily increase memory usage, particularly for large or complex repositories". After that first pass, the index is reused and updated in place, which is the difference between an index that pays a one-off cost and one that would pay it every session.

The scope is C++ in Copilot CLI, not every language and not every Copilot surface. A JavaScript, Python or Go project sees no change from this release, and a C++ project reached through a different tool sees no change either. The feature depends on the Microsoft C++ Language Server picking up the project's compilation information, so a build system that does not produce a compile_commands.json (or the equivalent metadata the language server expects) needs one before Copilot can index it.

Why cross-file matters here specifically

C++ is the language where the gap between "what the editor happens to have loaded" and "what the code actually references" is largest. A definition in one header pulls declarations from another, a template instantiation reaches into a fourth, and a reference search across a codebase means visiting more files than any editor keeps open at once. A Copilot answer that reasons only over open files is guessing at how those pieces fit together, and a plain grep on the tree beats it on symbol search whenever the symbol is not in the buffer.

The honest test for a team on C++ is a check against the workflow that was slowest without this: an unfamiliar codebase pulled fresh, a request to trace where a class is used, and a measurement of whether Copilot's list matches what a language-server "find references" returns. Every answer that does not match is one the index either did not build or the compilation information did not reach; every answer that does is one that used to depend on a file being open.

Source

GitHub Changelog, Faster C++ code intelligence with whole codebase indexing, 22 September 2026.