Use chat in VS Code

Use chat in Visual Studio Code to ask about your codebase or make edits across your project by using natural language. Chat can operate in different modes, optimized for your use case, from asking questions to making multi-file edits or starting an autonomous coding workflow.

You might want to use chat in VS Code when you need to:

• Understand code - "Explain how this authentication middleware works"
• Debug issues - "Why am I getting a null reference in this loop?"
• Get code suggestions - "Show me how to implement a binary search tree in Python"
• Optimize performance - "Help me improve the efficiency of this database query"
• Learn best practices - "What's the recommended way to handle errors in async functions?"
• Get VS Code tips - "How do I customize keyboard shortcuts?"

Prerequisites

• Install the latest version of Visual Studio Code
• Access to Copilot. Copilot Free plan and get a monthly limit of completions and chat interactions.

Access chat in VS Code

You can use natural language chat in different ways in VS Code, each optimized for a specific use case and task.

Experience	Use case	User experience
Chat view ⌃⌘I (Windows, Linux Ctrl+Alt+I)	Have an ongoing, multi-turn chat conversation in a dedicated view on the side. Switch between different chat modes to ask questions, make code edits across files, or start an autonomous coding workflow.
Inline chat ⌘I (Windows, Linux Ctrl+I)	Start a chat conversation directly from the editor (editor inline chat) or integrated terminal (terminal inline chat) to get suggestions in-place.
Quick Chat ⇧⌥⌘L (Windows, Linux Ctrl+Shift+Alt+L)	Ask a quick question and get back into what you were doing.

Access each chat experience by using the corresponding keyboard shortcuts or via the Copilot menu in the VS Code title bar.

Choose a chat mode

Chat modes are predefined configurations to customize chat in VS Code for specific tasks, such as asking questions, making code edits, or performing autonomous coding tasks. VS Code comes with three built-in chat modes: Ask, Edit, and Agent. You can also define your own chat modes for specific scenarios, such as planning a new feature, or researching implementation options.

To switch between chat modes, open the Chat view (⌃⌘I (Windows, Linux Ctrl+Alt+I)), and then select the desired mode from the chat mode dropdown list.

Learn more about chat modes in VS Code.

Change the language model

VS Code offers different built-in language models to choose from. Some models are optimized for fast coding tasks, while others are better suited for slower planning and reasoning tasks. Use the model picker in the chat input field to change the model that Copilot uses for generating a response.

You can also add models from other model providers (preview) and use them in chat. Get more details about how to use models from other providers.

Submit a chat prompt

Type a natural language prompt in the chat input box to make chat requests in VS Code. Depending on the chat mode, you can ask questions about your codebase, get code suggestions, or make code edits across multiple files in your project and invoke specialized tools.

A chat response might contain a combination of rich content such as Markdown text, code blocks, buttons, file trees, and more.

To get more relevant responses or reference specific files or artifacts in your workspace, such as test failures or terminal output, add context to your chat prompt by #-mentioning relevant context items.

Add chat context

VS Code tries to determine the intent and scope of the chat request based on your natural language prompt. To help get more relevant responses, provide additional context in your chat prompt, such as files, test results, terminal output, and more.

Use the Add Context button in the Chat view or type #-mentions to add context to your chat prompt. For example, type #codebase to perform a full codebase search, or #<file | folder | symbol> to refer to a specific file, folder, or symbol in your workspace. Type # in the chat input field to view the list of context items.

If you have installed MCP servers or extensions that provide tools, you can also directly reference these tools by #-mentioning them in your chat prompt. For example:

• summarize #fetch code.visualstudio.com/updates: retrieve the contents of a web page using the fetch built-in tool
• what is terminal suggest #githubRepo microsoft/code: perform a code search in a GitHub repository using the githubRepo built-in tool.

You can also directly reference an agent mode tool in your prompt by typing # followed by the tool name. You can do this in all chat modes (ask, edit, and agent mode). For example, use the #fetch tool to add the content of a web page as context to your chat prompt, or use #githubRepo to perform a code search in a GitHub repository.

Get more details about adding context to your chat prompt.

Prompt examples

For more prompt examples, see the Copilot Chat Cookbook in the GitHub documentation.

Vision

Chat supports vision capabilities, which means you can attach an image as context to your chat prompt and ask questions about it. For example, attach a screenshot of a block of code and ask to explain it, or attach a sketch of a UI and ask agent mode to implement it.

Chat history

As you iterate and send multiple chat prompts in a chat session, VS Code uses the history of chat prompts and responses as context for your current chat prompt. This means that you can ask follow-up questions or clarify your previous question without having to repeat the context. For example, you can ask "How does this differ from ...", "Now add a test case", "explain in more detail", and more.

At any time, you can create a new chat session by using the New Chat (+) button (⌘N (Windows, Linux Ctrl+N)) in the Chat view. This can be useful if you want to move to a different topic and avoid the previous context and history.

To view the history of chat sessions, select the Show Chats... button in the Chat view or by using the Chat: Show Chats... command in the Command Palette. Select a history entry to open that chat session in the Chat view and continue the conversation.

You can export all prompts and responses for a chat session in a JSON file with the Chat: Export Chat... command in the Command Palette. To copy a specific chat prompt or response to the clipboard in Markdown format, right-click the message and select Copy. To copy the entire chat session, right-click the Chat view and select Copy All.

Edit a previous chat request (Experimental)

You can edit a previous chat request in the active chat session. This is useful if you want to refine your prompt or correct a mistake. Editing a chat request is equivalent to reverting the request and then submitting a new request with the edited prompt.

When you edit a previous chat request, the following steps are performed:

1. The edited request and all subsequent requests and responses are removed from the conversation history.
2. Any edits that were made by these requests are reverted to their state before the request was made.
3. The edited request is added to the conversation history and submitted to the language model for a new response.

You can configure editing of previous chat request with the chat.editRequests setting:

• inline: Select the request in the Chat view to make it editable in-place. Use Escape to exit the edit mode.
• hover: Hover over a chat request and select the edit icon (pencil) to make it editable in-place. Use Escape to exit the edit mode.
• input: Hover over a chat request and select the edit icon (pencil) to edit the request in the chat input field.
• none: Disable editing of chat requests in the Chat view.

Revert chat requests with checkpoints (Preview)

Chat checkpoints provide a way to restore the state of your workspace to a previous point in time, and are particularly useful when chat interactions resulted in changes across multiple files.

When checkpoints are enabled, VS Code automatically creates snapshots of your files at key points during chat interactions, allowing you to return to a known good state if the changes made by chat requests are not what you expected or if you want to try a different approach.

To enable checkpoints, configure the chat.checkpoints.enabled setting.

Restore a checkpoint

When you restore a checkpoint, VS Code reverts the workspace to the state it was in at the time of that checkpoint. This means that all changes made to files after that checkpoint will be undone.

To restore your workspace to a previous checkpoint:

1. In the Chat view, navigate to previous chat request in the chat session.

2. Hover over the chat request and select Restore Checkpoint.

   

3. Confirm that you want to restore the checkpoint and undo any file changes made after that point.

   Notice that the chat request is removed from the conversation history, and the workspace files are restored to their state at the time of the checkpoint.

Redo after restoring

After restoring to a previous checkpoint, you can redo the changes that were undone. This might be useful if you inadvertently restored to a checkpoint.

To redo changes after restoring a checkpoint, select Redo in the Chat view.

View file changes in checkpoints

To help you understand the impact of each chat request and make it easier to decide which checkpoint to restore to, enable the chat.checkpoints.showFileChanges setting. This shows a list of files that were modified at the end of each chat request, along with the number of lines added and removed in each file.

Open chat in an editor tab or separate window

You can open a chat session as a separate editor tab, or even as a separate, floating window. This functionality enables you to have multiple chat sessions open at the same time.

In the Chat view, select the ... icon in the top-right corner, and then select Open Chat in Editor or Open Chat in New Window. The following screenshot shows the Chat view running in a floating window:

By default, the chat session opens in compact mode, which hides the title bar and other UI elements. Select the compact mode icon in the floating window title bar to toggle between compact and normal mode.

Optionally, enable the Always on Top mode to always keep the Chat view on top of other windows.

Learn more about floating windows in VS Code.

Use voice interactions

With the voice control capabilities in VS Code, provided by the VS Code Speech extension, you can initiate a chat conversation by using your voice:

• Use your voice to dictate your chat prompt
• Use the "Hey Code" voice command to start a voice session with Copilot Chat
• Accelerate voice input for chat by using the "hold to speak" mode

Learn more about how to use voice interactions in VS Code.

Chat Debug view

The Chat Debug view is a dedicated view where you can see the details of AI requests and responses. For each chat request, you can see the system prompt, user prompt, and details about the context that is sent to the language model. You can view the detailed response from the language model, and responses from tools that are invoked as part of the chat request.

To open the Chat Debug view, run the Developer: Show Chat Debug View command from the Command Palette (⇧⌘P (Windows, Linux Ctrl+Shift+P)).

Start chat from the command line

You can start a chat session directly from the command line by using the chat subcommand in the VS Code CLI. This enables you to open a chat session in your current working directory with a prompt you provide.

For example, the following command opens chat for the current directory and asks "Find and fix all untyped variables":

code chat Find and fix all untyped variables

The chat subcommand has the following command-line options:

• -m, --mode <mode>: The chat mode to use for the chat session. Available options: ask, edit, agent, or the identifier of a custom mode. Defaults to agent.
• -a, --add-file <path>: Add files as context to the chat session.
• --maximize: Maximize the chat session view.
• -r, --reuse-window: Use the last active window for the chat session.
• -n, --new-window: Open an empty window for the chat session.

The chat subcommand also supports piping input from stdin by passing - at the end of the command. For example:

python app.py | code chat why does it fail -

Privacy and transparency

To enable more workspace search features for private repositories, we require additional permissions. If we detect that we don't have these permissions already, we will ask for them at startup. Once granted, we'll securely store the session for the future.

Learn more about security, privacy, and transparency in the GitHub Copilot Trust Center.

Frequently asked questions

How do I choose between the different chat modes?

The different chat modes are optimized for different use cases:

• Use editor inline chat to ask questions or make edits directly in the active editor. This is useful for making code changes or asking questions that are scoped to the active file.

• Use ask mode to ask questions about your codebase or technology concepts. The response might include code suggestions, which you can apply manually and individually to your codebase. Changes are not automatically applied to your codebase.

• Use edit mode to directly apply edits across multiple files in your codebase based on your chat prompt. You provide the relevant context and files for your prompt.

• Use agent mode to start an autonomous coding workflow, whereby the AI autonomously determines the relevant context and files, determines which tasks need to be performed to complete the request. It then iterates independently to achieve the desired outcome, fixing issues as they come up. Agent mode can invoke tools from extensions or MCP servers to perform specialized tasks, such as running terminal commands, validating test cases, or accessing APIs.

Additional resources

You can read more about GitHub Copilot and how to use it in VS Code in the GitHub Copilot documentation.

Or check out the VS Code Copilot Series on YouTube, where you can find more introductory content and programming-specific videos for using Copilot with Python, C#, Java, PowerShell, C++, and more.

Next steps

• Use ask mode to ask questions about your code
• Use agent mode to start an autonomous coding session
• Use edit mode to make code edits across multiple files