GitHub Copilot key terms

The share of Copilot suggestions a developer inserts rather than dismisses. It is a usage signal, not a quality score, and varies widely by language, file type, and task.

A Copilot Chat mode where the model plans multi-step work, edits multiple files, runs terminal commands, and iterates on errors until the task completes or it needs your approval.

A code referencing feature that matches Copilot output against public code so you can see whether a suggestion resembles known licensed or public repository content before accepting it.

The fixed amount of text a model can consider at once, spanning your prompt, attached files, conversation history, and retrieved snippets. Content beyond it is truncated or summarized.

Inline, ghost-text suggestions generated as you type in the editor. The model predicts the next tokens based on surrounding code, open tabs, and repository context.

A Copilot feature that surfaces matches between generated code and indexed public repositories, letting organizations review matches and set whether suggestions with matches are blocked.

A repository or organization setting that prevents Copilot from using specified files or paths as context for completions and chat. It reduces exposure but does not fully block all Copilot behavior.

Deliberately selecting what the model sees: open relevant files, reference specific paths, summarize constraints, and remove noise so suggestions match your codebase conventions.

A conversational interface in the IDE, GitHub, and CLI that answers questions, explains code, proposes edits, and can act on the workspace with your approval.

Aggregated usage data such as active users, acceptance rates, and feature adoption, available through organization dashboards and APIs for measuring rollout impact.

Subscription levels such as Individual, Business, and Enterprise that differ in policy controls, audit logs, data retention options, and model access.

A repository or personal file of standing guidance, such as coding standards and preferred libraries, that Copilot applies automatically to chat and generation requests.

How long prompts and suggestions are stored. Policies differ by plan and setting, and organizations can often opt out of retention for model training.

A server-side filter that blocks suggestions closely matching public code, reducing the chance of inadvertently reproducing licensed snippets.

A Copilot Chat mode that proposes coordinated changes across selected files as a reviewable diff, without autonomously running commands or looping on its own.

The level at which content exclusions apply, such as organization or repository, and the precedence rules that determine which setting wins when both exist.

Including a small number of worked examples in the prompt so the model imitates the desired format, naming style, or pattern rather than guessing.

The dimmed inline suggestion shown in the editor before acceptance. It is a preview, not committed code, and can be cycled through with keyboard shortcuts.

Confident output that is wrong, such as a nonexistent API, package, or parameter. Review and test generated code rather than trusting plausibility.

The practice of reviewing, testing, and taking responsibility for Copilot output. The developer remains the author and is accountable for correctness, security, and licensing.

Predictions of the next place your code needs changing after an edit, offered as navigable suggestions rather than only completions at the cursor.

The instruction and supporting context you give Copilot. Its specificity, constraints, and examples strongly shape the relevance and correctness of the response.

Structuring requests with clear goals, constraints, output format, and relevant context so the model produces usable results with fewer correction cycles.

A reusable Markdown file of instructions that can be invoked in chat, letting teams standardize recurring tasks such as reviews, test generation, or documentation.

Chat syntax that attaches a specific file, symbol, or repository element to the prompt, giving the model precise context instead of relying on inferred workspace state.

A setting that suppresses suggestions matching public repositories above a similarity threshold, complementing code referencing and duplicate detection.

Building a searchable representation of a repository so chat can retrieve relevant files and symbols. Indexing scope interacts with content exclusions and permissions.

Fetching relevant snippets from the workspace or index to place in the prompt, which is how Copilot grounds answers in your code rather than only its training data.

Assigning, revoking, and auditing Copilot licenses across an organization, including policy assignment and ensuring only authorized users have access.

A shorthand in chat that triggers a built-in workflow, such as explaining code, generating tests, or fixing a problem, without writing a full prompt.

The act of inserting a Copilot suggestion. Accepting means you own the code, so review it for correctness, security, and style before committing.

Operational and usage data collected about Copilot features. Understand what is collected, where it flows, and how it differs from prompt retention.

Using Copilot to draft unit tests from a selected function or file, then reviewing edge cases and assertions rather than accepting them blindly.

The mechanism by which an agent invokes external capabilities such as search, terminal commands, or MCP servers, with approvals controlling what it may execute.

An open protocol for connecting AI assistants to external tools and data sources, letting Copilot access services configured by your organization or team.

Checking generated code through reading, compiling, running tests, and security review. Passing tests alone does not prove correctness or safety.

Asking the model to perform a task with instructions only, no worked examples. It works well for common tasks but is less reliable for strict formats.