Skip to content

GitHub Copilot: features, surfaces and plans

GitHub Copilot is GitHub’s AI assistant for writing, understanding, reviewing and shipping code. It started in 2021 as an autocomplete for the editor and has grown into a family of features: inline suggestions while you type, a chat that answers questions about your code, agents that work through multi-step tasks and open pull requests, and automated code review. Because it is built into GitHub, it can use the same repositories, issues and pull requests your team already works with.

This page is the starting point for the Copilot guides on this site. It explains what Copilot does, where you can use it, how the plans differ and how to start, with prompts based on an invented OrdersPortal application. Facts were checked against the GitHub Docs source in November 2026; Copilot changes monthly, so check the official plan page before you make a purchasing decision.

Topic What it answers
How Copilot works Assistive versus agentic help, and what context Copilot sees.
Copilot features Inline suggestions, chat, agents, code review, tests and security fixes.
Where you can use Copilot IDEs, GitHub.com, the CLI, the Copilot app and mobile.
Plans and access Free, individual and organization plans.
Getting started The first hour with Copilot.

GitHub’s documentation groups Copilot’s capabilities into four categories. Assistive features respond while you work: they suggest code, answer questions and explain code, and you review and apply each suggestion. Agentic features take a goal, research the repository, plan, edit files, run tools and prepare a pull request, while you remain responsible for reviewing and approving the result. Customizations such as repository instructions, prompt files, agent skills and MCP servers tell Copilot about your project’s conventions and tools. External agents and models from other providers can work alongside it.

Copilot builds a prompt from what is around you: the file you are editing, nearby and open files, your selection, the chat history, and, depending on the feature, indexed repository content, issues and pull requests. The prompt goes to a large language model; Copilot offers models from several providers, including OpenAI, Anthropic, Google and Microsoft, and which ones you can pick depends on your plan and your organization’s policies. The answer is a prediction, not a lookup, which is why every suggestion needs the same review as code from a new teammate.

An illustrative comment-to-code prompt (the suggestion differs between runs and models):

// Parse an OrdersPortal query string like "status=shipped&customer=C-12"
// into an object. Decode keys and values, ignore empty pairs,
// and keep the last value when a key repeats.
function parseOrderQuery(queryString) {
// Copilot proposes the body here as ghost text
}

Pitfalls:

  • Plausible is not correct. Models can invent APIs, miss edge cases or use outdated library versions. Run the tests and read the diff.
  • Context is limited. Copilot does not see your whole system. State constraints (types, frameworks, error handling) in the prompt or in repository instructions.

The feature set differs by plan and by client, but the core capabilities are:

  • Inline suggestions (“ghost text”): grey predictions of the next line or block as you type, accepted with Tab. In VS Code, next edit suggestions go further and propose the next related change, such as updating callers after you add a parameter. See Copilot in VS Code.
  • Copilot Chat: ask questions about code, frameworks and errors; request explanations, refactors and fixes; and generate unit tests for selected code. Chat is available in IDEs, on GitHub.com and in GitHub Mobile.
  • Agents: in the IDE, agent mode edits several files and runs commands and tests in a loop. On GitHub, you can assign an issue to the Copilot cloud agent, which works in a sandbox and opens a pull request for review. See GitHub Copilot Agent.
  • Copilot code review: request Copilot as a reviewer on a pull request; it leaves comments, often with suggested changes, which you accept or dismiss like any review.
  • Pull request and commit help: summaries of pull requests and generated commit messages.
  • Copilot CLI: the same assistant in the terminal. See GitHub Copilot CLI.
  • Copilot Autofix: suggested fixes for code scanning alerts. It is branded as a Copilot feature but requires a GitHub Code Security or GitHub Advanced Security license.

Example of reviewing a pull request with Copilot on GitHub.com (screenshots from the site’s earlier walkthrough):

Requesting a Copilot review on a pull request

Copilot review comments on a pull request

A test-generation prompt in chat, with the selection set to a pricing function:

Write xUnit tests for OrderPricing.CalculateTotal.
Cover: empty order (total 0), one line, a 10% discount code,
a discount larger than the subtotal (total must not go below 0),
and rounding to 2 decimal places. Use the existing OrderBuilder test helper.

Pitfalls:

  • Generated tests can assert the bug. Tests written from the implementation confirm what the code does, not what it should do. Give the expected behavior in the prompt.
  • Agents need guardrails. Agent output goes through the same branch protection, required reviews and CI as human work; do not exempt it.

According to GitHub’s documentation in November 2026, Copilot is available in:

Surface Good for
IDEs: VS Code, Visual Studio, JetBrains IDEs, Eclipse and Xcode (plus Vim and Neovim for inline suggestions) Inline suggestions, chat about nearby code, agent mode
GitHub website Questions about repositories, issues and pull requests; assigning work to the cloud agent; code review
GitHub Copilot CLI Terminal workflows and agentic tasks from the command line
GitHub Copilot app Coordinating agent work across tasks and repositories
GitHub Mobile Quick questions away from the desk
GitHub Desktop Commit messages and summaries

Settings such as your plan, custom instructions, agent skills and connected MCP servers apply across surfaces where supported, so you configure them once. To see which features your account has, open https://github.com/settings/copilot/features.

Guides on this site: Copilot in VS Code, Copilot in Visual Studio 2022, Copilot CLI and GitHub MCP server.

Pitfalls:

  • Feature parity varies. A feature announced for VS Code may arrive later, or never, in another IDE. Check the Copilot feature matrix before standardizing a workflow.
  • Organization policy wins. If a feature is missing, your organization may have disabled it; ask an administrator rather than switching accounts.

GitHub offers individual plans (Copilot Free, Copilot Student, Copilot Pro, Copilot Pro+ and Copilot Max) and organization plans (Copilot Business and Copilot Enterprise). Usage beyond inline suggestions is measured in AI credits: each plan includes a monthly allowance, and paid plans add access to more and stronger models. List prices in the GitHub Docs in November 2026:

Plan Price Notes
Copilot Free $0 Limited allowance; automatic model selection only
Copilot Student $0 For verified students
Copilot Pro $10 per month Free for some users, such as eligible teachers and open source maintainers
Copilot Pro+ $39 per month Larger allowance, premium models
Copilot Max $100 per month Highest individual allowance
Copilot Business $19 per granted seat per month Central management and policy control
Copilot Enterprise $39 per granted seat per month Requires GitHub Enterprise Cloud; larger credit pool

For members of an organization, an administrator must assign a seat before you can use Copilot at the organization’s expense. Copilot is not available for GitHub Enterprise Server.

Pitfalls:

  • Allowances change. Credit amounts and model multipliers are revised often; read the current plans page instead of relying on a screenshot.
  • Personal and work accounts are separate. Sign in with the account that holds the seat, especially in an editor shared between work and personal projects.
  1. Get access: sign up for Copilot Free or a paid plan at github.com/copilot, or request a seat from your organization at github.com/settings/copilot.
  2. Set up your editor. In VS Code, Copilot is built in: hover over the Copilot icon in the Status Bar and select Use AI Features, then sign in. Details are in Copilot in VS Code.
  3. Write a descriptive comment or function signature and watch for ghost text; press Tab to accept.
  4. Open chat and ask about code you already know, so you can judge the answers: “Explain how OrderPricing applies discounts and list edge cases that are not tested.”
  5. Add a .github/copilot-instructions.md file describing the stack and conventions, so every answer starts from the same context. See Copilot best practices.

What is the difference between assistive and agentic Copilot features? Assistive features (inline suggestions, chat answers, explanations) respond to you and you apply each suggestion. Agentic features take a goal, plan, edit several files, run tools and return a change, usually a pull request, which you review. Responsibility for the result stays with the developer in both cases.

What context does Copilot use for an inline suggestion? The code around the cursor, the current file and related open files, plus repository context and custom instructions where the client supports them. It does not see your whole system, which is why explicit constraints in comments and instructions matter.

How do you keep sensitive files out of Copilot? Organization and repository administrators on Business and Enterprise can configure content exclusion, which stops excluded files from being used for suggestions, chat context and code review. It is configured by path pattern and has documented limitations, for example symlinks.

Senior scenario: your company is rolling out Copilot Business to 300 developers. What do you put in place first? Policies before seats: decide which features and models are enabled, configure content exclusion for secrets and regulated code, and set the public code matching policy. Keep branch protection, required reviews and CI unchanged for AI-assisted and agent pull requests. Add repository instructions to the main repositories, run a pilot with a few teams, and measure outcomes (review time, defect rates, usage metrics) before a full rollout.