Does anyone know what will be sent to LLM when I use Github copilot?

Files?
Prompts?
Code snippets?

Comments

[Ashley Childress]

Short answer — yes. 😆 If you want all the fine print, GitHub has a solid breakdown on their trust page.

In practice, what gets sent depends on your IDE and settings. By default, Copilot includes any open files (with names and snippets) plus your prompts, passed to the model as user messages. Custom instructions are always included too, unless you’ve changed that or are running in a specific chat mode.

For exclusions, you’ve got a couple of options: a .copilotignore file that works just like .gitignore (glob patterns, nested, or directory-level) and policies that can be set at the repo or org level. And importantly, GitHub enforces a strict no-training policy—none of your data is used to train their models.

Hope that clears it up 😀

[berry22jelly]

Is Github copilot using embedding-based semantic search on my codebase? They are store locally or on cloud?

[Ashley Childress]

I know they have what they call "indexing" stored both locally and in the cloud. Now, what that mean exactly? I have no idea. You can probably find the info on their trust page somewhere though if you were to dig enough for it.