What is the GitHub Copilot app?
The GitHub Copilot app is GitHub's desktop application for running AI coding agents: you connect a repository, start several agent sessions in parallel, each on its own branch and git worktree, and take each one from issue to merged pull request in the same window. It runs on macOS, Windows and Linux and is built on GitHub Copilot CLI.
The name causes confusion, so to be clear: this is not the GitHub Mobile app and not the Copilot extension inside VS Code. It is a separate desktop client.
The timeline from GitHub's changelog:
| Date | Milestone |
|---|---|
| 14 May 2026 | Technical preview |
| 17 June 2026 | Generally available for macOS, Windows and Linux |
| 7 July 2026 | Available on every Copilot plan, including Copilot Free and GitHub Education |
According to the GitHub Copilot app docs, the main features are parallel workspaces, three session modes, a model picker with bring your own key, GitHub issue and pull request handling, custom instructions, MCP servers and agent skills, scheduled automations, chats, canvases, and computer use for local sessions on macOS and Windows.
How do you download it and sign in on Mac, Windows or Linux?
Download it from GitHub's Copilot app page. The builds come from GitHub's releases: a .dmg for Apple silicon and Intel Macs, a setup .exe for Windows x64 and Arm, and an .AppImage for Linux x64. You also need a GitHub account and Git installed.
Then, per the quickstart:
- Open the app and click Sign in to GitHub.
- On GitHub Enterprise Cloud with data residency, choose Use GitHub Enterprise and enter your
*.ghe.comhostname. GitHub Enterprise Server is not supported. - Without a Copilot plan, choose to sign up for one or continue with your own model provider.
- Pick repositories from your recent activity, add a local folder, or skip.
- Choose a theme and finish onboarding.
To add more later, click + next to Sessions and choose Open local folder, Add GitHub repository or Clone repository (for Git URLs outside GitHub, such as Azure DevOps).
On Copilot Business or Enterprise, an admin must leave the GitHub Copilot app policy enabled. The docs say it is on by default and separate from the Copilot CLI policy. The June and July changelog posts instead said admins must have the Copilot CLI enabled in policy settings, so the current docs reflect a later change. If the app is unavailable on a work account, ask your admin about the app policy.
How do sessions, worktrees and cloud sandboxes work?
Each session runs in its own isolated workspace, so parallel tasks do not collide. When you start one, you pick where it runs:
| Location | What it means |
|---|---|
| New working tree | A dedicated git worktree and branch on your machine |
| Local repository | Works directly in your checked-out repo |
| Cloud sandbox | A fully isolated environment hosted by GitHub (public preview) |
You can also turn on a local sandbox, which keeps the session on your machine but restricts the filesystem, network and credential access of agent-run tools. Set a default per repository or type /sandbox on or /sandbox off in a session.
Three settings shape how the agent behaves, and you can change each mid-session:
- Session mode: Interactive (works with you and waits for input), Plan (writes a plan for you to approve) or Autopilot (fully autonomous).
- Tool permissions: Ask every time, Assisted (a model approves requests it judges safe and asks about the rest) or Approve all.
- Model and reasoning effort: pick a model, or Auto to let the app choose by task complexity.
In the prompt box, # references an issue, @ adds a file and / opens commands. Useful ones from the slash command reference include /plan, /autopilot, /fleet (several agents on one task), /fork, /review, /security-review (public preview), /rubber-duck (critique from a different model), /pr-open, /pr-fix-checks, /compact and /usage. For brainstorming without a branch, use Chats in the sidebar.
How do skills, MCP servers and custom instructions work in the app?
The app reuses what you already have. The customization docs say: "Any skills configured for your repositories or Copilot CLI are automatically available in the GitHub Copilot app." The same goes for MCP servers.
Skills. Copilot reads project skills from .github/skills, .claude/skills or .agents/skills in a repo, and personal skills from ~/.copilot/skills or ~/.agents/skills, though the docs note supported locations depend on the surface. To browse or add skills in the app, click Customize, then Skills. Run /skills reload to pick up a new skill mid-session. The Copilot agent skills guide covers the SKILL.md format.
The app also ships built-in skills, some run as slash commands:
| Skill | What it does |
|---|---|
/orchestrate |
Coordinates work across sessions and repositories with child sessions |
/pr-stack |
Builds a stack of dependent pull requests, one child session per layer |
/create-canvas |
Creates and troubleshoots canvas extensions in the side panel |
/af |
Searches Agent Finder for installable MCP servers, tools, skills and agents |
agent-merge |
Runs automatically to handle review comments, failing checks and conflicts when agent merge is on |
customize-cloud-agent |
Runs when you ask to set up the cloud agent environment |
MCP servers. Click Customize, then MCP, to browse featured or trending servers or add a custom one. See the Copilot MCP guide for configuration.
Instructions. Global instructions live in app settings under Sessions, in "App instructions", and apply to every session. Repository instructions live in app settings under Repositories. /init generates or improves repository instructions. The custom instructions guide covers instruction files.
Plugins and agents. Customize also lists plugins from configured marketplaces (packages of skills, hooks, custom agents, MCP servers and canvases), and /agent picks a custom agent.
Which plans include the app, and can you use your own model key?
The docs say the app "is available for all Copilot plans." They also give tips for using AI credits efficiently in the app, such as matching the model to the task and starting a new session when you switch tasks, and /chronicle cost-tips points out expensive patterns. /usage shows usage and rate-limit details for your plan. The Copilot pricing guide explains plans and AI credits.
You do not need a plan at all if you bring your own key. Adding LLM models to the app (public preview) supports OpenAI, Azure OpenAI, Microsoft Foundry, Anthropic, Ollama, Foundry Local, LM Studio and any OpenAI-compatible HTTP endpoint. To add one:
- Open app settings and click Model providers.
- Click Add provider and choose the provider.
- Enter the details, such as display name, base URL and API key.
- Click Add provider to save.
The models then appear in the picker next to GitHub-hosted ones, and credentials are stored in your system credential store. You still sign in with a GitHub account.
How is the app different from Copilot CLI, VS Code and the cloud agent?
All four can run Copilot as an agent. They differ in where the work happens and what surrounds it.
| GitHub Copilot app | Copilot CLI | VS Code agent mode | Copilot cloud agent | |
|---|---|---|---|---|
| Interface | Desktop app | Terminal | Your editor | GitHub.com, from issues and PRs |
| Runs on | Your machine or a GitHub cloud sandbox | Your machine | Your machine | An ephemeral cloud environment |
| Best for | Directing several tasks at once and landing PRs | Scripting and terminal-first work | Editing alongside the agent | Delegating an issue and reviewing the PR later |
Because the app is built on the CLI, they share skills, MCP servers and session history: /chronicle standup summarizes work from both. If you live in the terminal, the CLI is enough. If you juggle several branches at once, the app saves window switching. The cloud agent guide covers assigning issues on GitHub.com.
How do you keep your skills in sync across the app and other agents?
Since Copilot reads project skills from .claude/skills and .agents/skills as well as .github/skills, one set of SKILL.md folders can serve Copilot and Claude Code. Kipt keeps prompts and Agent Skills as Markdown in a private GitHub repo you own, and npx skills add https://github.com/<you>/<repo>.git --all -g installs every skill in one step on a new machine.
Sources: GitHub Copilot app, Getting started with the GitHub Copilot app, Working with agent sessions, Built-in skills and the GitHub changelog posts linked above.