GitHub Copilot capabilities

Synchronous, autonomous agent inside the IDE that plans multi-step coding tasks, edits files across the workspace, runs commands, and iterates until the task is complete.

Copilot can review pull requests on github.com and review uncommitted changes in the IDE, leaving inline review comments and suggesting fixes.

Terminal-native agentic assistant; natural-language conversation in the shell that proposes and executes commands and code edits with per-action approval by default (an opt-in autopilot mode runs without approval stops). Generally available since 2026-02-25.

Conversational AI panel and inline chat for asking coding questions, generating code, and explaining selections. Includes slash commands such as /explain, /fix, /tests, /help, and (surface-specific) /doc, /new, /optimize, /simplify, /fixTestFailure.

Persistent context that lets Copilot agents discover, store, and reuse facts about a codebase (coding conventions, architectural decisions, build commands, cross-file dependencies) across interactions, so users do not have to re-supply that context each time. Stores two kinds of entries: repository-level facts (available to anyone with access to that repository) and user-level preferences (implied or stated personal preferences, private to the individual user across repositories). As of March 4, 2026 it is on by default for Copilot Pro and Pro+ individual users in public preview; users can disable it in personal Copilot settings.

A multi-language SDK, generally available since 2026-06-02, that gives developers direct programmatic access to the same agent runtime behind GitHub Copilot - planning, tool invocation, file edits, streaming, and multi-turn sessions - for embedding into third-party applications, services, and developer tools without building a custom orchestration layer.

Secure, shareable containers that organize context for Copilot Chat - bundling repositories, code, pull requests, issues, free-text content (transcripts, notes), images, file uploads, and custom instructions to ground Copilot answers in project-specific knowledge. Replaces the retired Copilot knowledge bases (sunset 2025-11-01).

Asynchronous cloud-hosted agent that, given an issue or task, researches the repo, drafts an implementation plan, makes changes on a branch in its own ephemeral GitHub Actions dev environment (running tests and linters), and opens a pull request you can review.

Iterative, multi-file edit session driven by chat prompts. Copilot proposes inline diffs in a 'working set' of files that you accept or reject individually or collectively.

A standalone desktop application (macOS, Windows, Linux) that serves as an agent-native home for directing AI coding agents across parallel workstreams, distinct from the in-editor Copilot experience. It centralizes starting sessions from GitHub issues/PRs/prompts, reviewing agent progress, and managing the full PR lifecycle without switching between terminal, IDE, and browser. Reached general availability on 2026-06-17 after an April 2026 preview and a June 2026-06-02 expanded technical-preview stage.

Natural-language full-stack app builder that extends Copilot from code-level help to app-level scaffolding and deployment. You describe an app in plain language and an AI agent generates a working full-stack web app - frontend, backend, AI features, and a database - with GitHub-account sign-in built in. You refine it via natural language, clickable visual controls, or by editing the code with Copilot, see changes in a live preview, create a GitHub repo in one click, and publish to secure hosting in one click.

First-party extensions for the major developer editors: VS Code, Visual Studio, JetBrains suite (IntelliJ, PyCharm, GoLand, WebStorm, Rider, etc.), Neovim/Vim, Xcode, and Eclipse.

Real-time, in-editor 'ghost text' code suggestions that appear at the cursor while you type; press Tab to accept. The core single-line and multi-line completion experience available across all supported editors.

Model Context Protocol (MCP) servers can be configured to extend Copilot Chat and Agent with custom tools and data sources. MCP is the only third-party capability hook - the GitHub App-based Copilot Extensions (and their skillsets) were sunset on 2025-11-10 and replaced by MCP.

User-selectable models in the Chat/Edit/Agent model picker. As of 2026-06 the picker spans OpenAI (GPT-5 mini, GPT-5.3-Codex, GPT-5.4, GPT-5.4 mini, GPT-5.4 nano, GPT-5.5), Anthropic Claude (Haiku 4.5, Sonnet 4.5/4.6, Opus 4.5/4.6/4.7/4.8 plus Opus 4.6 fast mode in preview, and Fable 5), Google Gemini (2.5 Pro, 3 Flash preview, 3.1 Pro preview, 3.5 Flash), Microsoft MAI-Code-1-Flash, and GitHub Raptor mini (preview).

Predicts both the location and content of the next likely edit based on edits you've already made, letting you tab through a chain of related changes across the file.

Seven plan tiers: Copilot Free, Copilot Student (free, eligibility-gated), Copilot Pro ($10/mo), Copilot Pro+ ($39/mo), Copilot Max ($100/mo), Copilot Business ($19/seat/mo), Copilot Enterprise ($39/seat/mo). Copilot Max was added 2026-06-01 as the top individual tier; individual-plan self-serve sign-ups reopened gradually from 2026-06-17.

Per-repository Markdown instruction files that prepend guidance to Chat/Edit/Agent requests. Repository-wide instructions live in .github/copilot-instructions.md; path-specific instructions in .github/instructions/*.instructions.md (with an applyTo glob in frontmatter). Reusable prompt files use the .prompt.md extension (in .github/prompts/). Agent-instruction files (AGENTS.md, plus CLAUDE.md and GEMINI.md) are also recognized on supported surfaces.