How to Use Copilot Chat with Files, Folders, Symbols, and Terminal Context

Copilot Chat context in VS Code can include your active file, current selection, and file name automatically. When the AI must consider other files, folders, symbols, terminal output, or web content, add them explicitly.

The techniques on this page work in both the Chat view and the /vscode/how-to-use-coding-agents-and-the-agents-window-in-vs-code.

#-mentions: add files, folders, and more

Type # in the chat input box. A menu appears showing the types of context you can add. Pick one, then select the specific item.

Common context choices

After you select a context type, VS Code shows a quick pick list of matching items. Start typing to filter.

Reference a file

Type #, start entering the file name, and select the matching file. You can also select Add Context and choose Files & Folders. The selected file appears as a context item in the chat input.

You can add multiple files to the same prompt. Each file increases the token count, so the context window indicator in the chat input fills up as you add more.

Reference a code symbol

To add a specific function, class, or variable, first open the file that contains it. Type # and select the symbol, or choose Add Context > Symbols and select it from the picker.

This is useful when you want to ask about one function in a large file without adding the entire file.

Reference terminal output

Select text in the integrated terminal, then type #terminalSelection in your chat prompt. The selected terminal text is added as context. This is useful for pasting error messages or command output into a chat question.

Terminal output can contain tokens, environment values, private paths, or customer data. Check the selection and remove sensitive text before sending the prompt.

Reference the entire codebase

Type #codebase to tell Copilot to search your entire workspace for relevant context. This uses VS Code's workspace indexing to find related files, symbols, and patterns. Use it when your question spans multiple files and you are not sure which files are relevant.

Reference web content

Include a URL directly or use #fetch with the URL to retrieve web content. For example, Summarize the relevant API changes #fetch https://api.example.com/changelog asks the AI to read that page before answering.

VS Code prompts for confirmation before accessing external URLs. The content is cached for a short time to avoid repeated network requests.

Faster ways to add context

Drag and drop

Drag files or folders from the Explorer view, Search view, or editor tabs directly into the chat input. Each dropped item becomes a context chip.

Add Context button

Select Add Context in the Chat view. A picker opens with available context categories, including Files & Folders and Symbols.

Attach images

Attach or drag an image into the Chat view to use vision capabilities. For example, add a UI mockup and ask for an implementation plan. Remove account details, tokens, notifications, and private paths from screenshots first.

@-mentions: choose who responds

Type @ in the chat input to invoke a chat participant. Participants are specialized assistants for specific domains.

Extensions can also contribute their own chat participants.

Monitor context usage

The chat input shows a context window indicator: a small bar that fills up as you add context. Hover over it to see:

• Token usage: the current token count out of the model's total context window (for example, 15K out of 128K).
• Credit cost: the total AI credits consumed in this session.
• Usage breakdown: how many tokens each context category uses.

When the bar fills up, VS Code automatically compacts the conversation by summarizing earlier messages. This is called context compaction and it happens in the background so you can keep chatting.

Manually compact the conversation

Type /compact in the chat input to trigger compaction on demand. Optionally add instructions: /compact focus on the database schema decisions. This is useful when the conversation has drifted and you want to refocus the AI on the most important parts.

Combine context for complex questions

A strong prompt combines multiple context types. For example:

Why does the login function fail when #terminalSelection shows "token expired"? Use the attached auth.ts file and validateToken symbol, then suggest a fix without changing files.

This prompt combines a file, terminal output, and a specific function while setting a safe read-only boundary. It gives the AI more useful evidence than a vague "fix my login" request.

For writing project-wide conventions that apply to every chat automatically, see /vscode/how-to-use-custom-instructions-in-vs-code-with-agents-md-and-instructions-md.