GitHub Copilot App Tour
Time: ~10 minutes
Copilot Features: Sessions, Chats, My Work, Automations, Customize, Search
This workshop is written for the GitHub Copilot app — the desktop app where you run agent sessions against your repositories. This page is a quick orientation so the later exercises make sense. If you already use the app daily, skim it and move on.
Follow the Copilot onboarding guide to install the app and sign in, then come back here.
Where things live
| Surface | What it's for |
|---|---|
| Sessions | Agent work on a repository. Choose an isolated worktree for parallel local work, your existing checkout for direct work, or a cloud sandbox when available. |
| Chats | Lightweight conversations that aren't tied to changing a repository — questions, research, drafting. |
| My Work | Browse and filter issues and pull requests, inspect CI and review activity, and start a session with the work item already in context. |
| Automations | Saved agent tasks that run manually, on a schedule, or in response to issue and pull request events. |
| Customize | Discover and manage plugins, skills, MCP servers, agents, and canvases. |
| Search | Search across your connected repositories from the app. |
You'll use Sessions for most of this workshop, Customize when you add MCP servers and custom agents, and My Work when you review the pull requests you produce.
Sessions run in isolated worktrees
For this workshop, start each session in a new worktree — a separate checkout on its own branch. The app can also work directly in an existing checkout or use a cloud sandbox, but a worktree is the safest default for parallel local exercises.
This matters for the workshop:
- You can run several exercises at once without one agent overwriting another's files.
- Your main checkout stays clean, so an experiment that goes sideways costs you nothing.
- Each session's changes land on a branch that is ready to become a pull request.
Tip: Start a new session per exercise. Fresh context produces better results than one long session carrying unrelated history.
Choose a mode
Every session runs in one of three modes. You can switch mid-session.
| Mode | Behavior | Use it when |
|---|---|---|
| Interactive | The agent works and checks in with you as it goes. | Most work. This is the default choice. |
| Plan | The agent researches, asks clarifying questions, and produces a plan for you to approve before writing code. | Ambiguous or multi-file work — like the cart feature in the next exercise. |
| Autopilot | The agent runs with minimal interruption until it's done. | Well-scoped, low-risk work you've already reasoned about. |
The workshop's rhythm is Plan → approve → Interactive: agree on the approach first, then let the agent build it.
Choose a model
Leave the model on Auto. It routes your request to a suitable model, and it's the right default for nearly everything in this workshop.
Switch to a higher-reasoning model only when you hit genuinely hard work — a complex plan spanning many files, or a bug you've already failed to fix once. Higher-reasoning models are slower and cost more, so use them deliberately rather than by habit.
See the model comparison for current options.
Warm up
Try this before the first exercise so the interface isn't new when it matters:
- Open the app and add your copy of the demo repository as a project.
- Start a new session on it.
- Leave the model on Auto and stay in Interactive mode.
- Prompt:
Give me a tour of this repository: the stack, how the frontend and API are organized, and how to run it locally.
- Note where the session's worktree and branch are shown — you'll come back to them when you open a pull request.
Next Steps
Continue to Feature Development to build your first feature.