Question

A backend developer at a fintech company is writing a completion in a private repository. They notice Copilot's completion service constructs a prompt before sending it to the model. During a security review, a colleague asks how Copilot decides which snippets from the surrounding code to include in that prompt, and what happens to those snippets after a suggestion is generated. Which statement most accurately describes the prompt-building and data-flow behavior for GitHub Copilot code completions?

Accepted Answer

Copilot uses relevance heuristics (such as the current file, cursor position, and content from related open files) to assemble the prompt locally, sends it over an encrypted connection, and for Copilot Business/Enterprise the prompt is not retained after the suggestion is returned.

Answer

Copilot uploads the entire repository to the model provider so it can index all files, then permanently stores the resulting embeddings to improve future suggestions for that repository.

Answer

Copilot builds the prompt on GitHub's servers by cloning the repository at request time, and always retains the full prompt for 28 days regardless of subscription tier.

Answer

Copilot only ever sends the exact line where the cursor sits, with no surrounding context, and caches every prompt indefinitely to guarantee deterministic suggestions.