[go: up one dir, main page]

copilot

Subscribe to all “copilot” posts via RSS or follow GitHub Changelog on Twitter to stay updated on everything we ship.

~ cd github-changelog
~/github-changelog|main git log main
showing all changes successfully

Today, we released an upgrade to the GPT-4o Copilot preview model for code completion.

What’s changing

The GPT-4o Copilot model released last month has been upgraded through reinforcement learning to deliver even more useful code completions.

Availability

This upgraded model is available today in public preview under the same gpt-4o-copilot name, with general availability expected in mid-April 2025.

Getting ready

To get started in Visual Studio Code or JetBrains, follow the instructions for your preferred IDE.

If you’re a Copilot Business or Enterprise user, your administrator first needs to enable this model for your organization by opting in to Editor preview features in the Copilot policy settings on github.com.

See more

Personal instructions in github.com

Personal instructions are now generally available for Copilot Chat on github.com! This means you can provide Copilot with important details about your preferences, such as your preferred language, response style, or even code standards.

To get started, open up Copilot Chat, click ..., and select Personal instructions. That’s it! Copilot will now incorporate your preferences for all chats in github.com.

💡 Looking for ideas? Here are some examples to kick things off:

  • Frameworks: “Use Vue 3 with the composition API.”
  • Language: “Always respond in Portuguese.”
  • Response preferences: “Each message should communicate a singular idea. That idea should be limited to ONE codeblock or paragraph.”
  • Code style and best practices: “Optimize code for readability and performance. Utilize SOLID principles.”
See more

Copilot features can go through different early access stages, which are typically enabled and configured through settings. The possible stages are as follows:

  • Experimental: This feature is still in development and not yet ready for general use.
  • Preview: This feature is still under refinement, but is ready to use. You’re welcome to provide feedback!
  • Stable: This feature is ready for general use.

Copilot Edits

Experimental agent mode improvements

Last month, we introduced agent mode for Copilot Edits in VS Code Insiders. In agent mode, Copilot can automatically search your workspace for relevant context, edit files, check them for errors, and run terminal commands (with your permission) to complete a task end-to-end.

Agent mode is available today in VS Code Insiders, and we just started rolling it out gradually in VS Code Stable. Once you have agent mode enabled, you’ll see a mode dropdown in the Copilot Edits view — simply select Agent.

We made several improvements to the UX of tool usages this month:

  • You can keep track of which commands were run since terminal commands are now shown inline.
  • You can edit the suggested terminal command in the chat response before running it.
  • You can confirm a terminal command with the Ctrl + Enter shortcut.

Agent mode autonomously searches your codebase for relevant context. Expand the message to see the results of which searches were done.

Screenshot that shows the expandable list of search results in Copilot Edits.

We’ve also made various improvements to the prompt and behavior of agent mode:

  • The undo and redo actions in chat now undo or redo the last file edit made in a chat response. This is useful for agent mode, as you can now undo certain steps the model took without rolling back the entire chat response.
  • Agent mode can now run your build tasks automatically or when instructed to do so. You can disable this functionality by using the VS Code setting github.copilot.chat.agent.runTasks if the model runs tasks when it shouldn’t.

Learn more about Copilot Edits agent mode or read the agent mode announcement blog post.

If you are a Copilot Business or Enterprise user, an administrator of your organization must opt in to “Editor Preview Features” for agent mode to be available.

Notebook support in Copilot Edits in preview

You can now use Copilot to edit notebook files with the same intuitive experience as editing code files. Create new notebooks from scratch, modify content across multiple cells, insert and delete cells, and change cell types. This preview feature provides a seamless workflow when working with data science or documentation notebooks.

We recommend using VS Code Insiders and the pre-release version of GitHub Copilot Chat to get the best notebook editing experience with Copilot.

Refined editor integration

We have polished the integration of Copilot Edits with code and notebook editors:

  • The viewport remains in place, making it easier to focus on what changes while changes are being applied.
  • The edit review actions are now “Keep” and “Undo” to better reflect what’s happening. Copilot Edits applies and saves these changes as users keep or undo them.
  • The next file is automatically revealed after you keep or undo changes to a file.

Refreshed UI

Files that are attached and not yet sent are now rendered as regular chat attachments. Only files that have been modified with AI are added to the changed files list.

You can use the VS Code setting chat.renderRelatedFiles to enable getting suggestions for related files. Related file suggestions are rendered below the chat attachments.

Screenshot that shows the updated Copilot Edits attachments and changed files user experience.

Removed Copilot Edits limits

We removed the limit on the number of files you can attach to your Copilot Edits prompt and the client-side rate limit.

Note that service-side usage rate limits still apply.

Custom instructions generally available

Custom instructions enable you to tailor GitHub Copilot to provide chat responses and code suggestions to the way you and your team work. Describe your specific requirements in a .github/copilot-instructions.md file in your workspace and enable them with the VS Code setting github.copilot.chat.codeGeneration.useInstructionFiles.

Custom instructions are generally available with .github/copilot-instructions.md. Make sure that the github.copilot.chat.codeGeneration.useInstructionFiles setting is enabled to have Copilot use these instructions when generating responses.

Learn more about custom instructions in Copilot.

Smoother authentication flows in chat

If you host your source code in a GitHub repository, you can leverage several features, including advanced code searching, the @github chat participant, and more!

However, for private GitHub repositories, VS Code needs to have permission to interact with your repositories on GitHub. For a while, this was presented with our usual VS Code authentication flow, where a modal dialog showed up when you invoked certain functionality (for example, asking @workspace or @github a question, or using the #codebase tool).

To make this experience smoother, we’ve introduced this confirmation in chat:

Screenshot that shows the authentication confirmation dialog in Chat, showing the three options to continue.

Not only is it not as jarring as a modal dialog, but it also has new functionality:

  1. Grant: you’re taken through the regular authentication flow like before (via the modal).
  2. Not Now: VS Code remembers your choice and won’t bother you again until your next VS Code window session. The only exception to this is if the feature needs this additional permission to function, like @github.
  3. Never Ask Again: VS Code remembers your choice and records it in the VS Code setting github.copilot.advanced.authPermissions. Any feature that needs this additional permission will fail.

It’s important to note that this confirmation does not confirm or deny Copilot (the service) access to your repositories. This is only how VS Code’s Copilot experience authenticates. To configure what Copilot can access, please read the docs on content exclusion.

More advanced codebase search in Copilot Chat

When you add #codebase to your Copilot Chat query, Copilot helps you find relevant code in your workspace for your chat prompt. #codebase can now run tools like text search and file search to pull in additional context from your workspace.

Set the github.copilot.chat.codesearch.enabled VS Code setting to enable this behavior. The full list of tools is:

  • Embeddings-based semantic search
  • Text search
  • File search
  • Git modified files
  • Project structure
  • Read file
  • Read directory
  • Workspace symbol search

Attach problems as chat context

To help with fixing code or other issues in your workspace, you can now attach problems from the Problems panel to your chat as context for your prompt.

Either drag an item from the Problems panel onto the Chat view, type #problems in your prompt, or select the paperclip 📎 button. You can attach specific problems, all problems in a file, or all files in your codebase.

Attach folders as context

Previously, you could attach folders as context by using drag and drop from the Explorer view. Now, you can also attach a folder by selecting the paperclip 📎 icon or by typing #folder: followed by the folder name in your chat prompt.

Collapsed mode for Next Edit Suggestions in preview

We’ve added a collapsed mode for NES. When you enable this mode, only the NES suggestion indicator is shown in the left editor margin. The code suggestion itself is revealed only when you navigate to it by pressing Tab. Consecutive suggestions are shown immediately until a suggestion is not accepted.

The collapsed mode is disabled by default and can be enabled by configuring editor.inlineSuggest.edits.showCollapsed:true, or you can toggle it in the NES gutter indicator menu.

Screenshot that shows the Next Edit Suggestions context menu in the editor left margin, highlighting the Show Collapsed option.

Change completions model

You can now change the model providing code completions by using the Change Completions Model command in the Command Palette or the Configure Code Completions item in the Copilot menu in the title bar.

If you’re a Copilot Business or Enterprise user, your Administrator needs to enable certain models for your organization by opting in to Editor Preview Features in the Copilot policy settings on github.com.

Model availability

This release, we added more models to choose from when using Copilot. The following models are now available in the model picker in Visual Studio Code and Copilot Chat on github.com:

  • GPT 4.5 (Preview): OpenAI’s latest model, GPT-4.5, is now available in GitHub Copilot Chat to Copilot Enterprise users. GPT-4.5 is a large language model designed with advanced capabilities in intuition, writing style, and broad knowledge. Learn more about the GPT-4.5 model availability in the GitHub blog post.
  • Claude 3.7 Sonnet: Claude 3.7 Sonnet is now in preview for all customers on paid Copilot plans. This new Sonnet model supports both thinking and non-thinking modes in Copilot. In initial testing, we’ve seen particularly strong improvements in agentic scenarios. Learn more about the Claude 3.7 Sonnet model availability in the GitHub blog post.

Copilot Vision in preview

We’re quickly rolling out end-to-end Copilot Vision support in this version of Copilot Chat. This lets you attach images and interact with images in chat prompts. For example, if you encounter an error while debugging, attach a screenshot of VS Code, and ask Copilot to help you resolve the issue. You could also use it to attach a UI mockup and let Copilot provide some HTML and CSS to implement the mockup.

Animation that shows an attached image in a Copilot Chat prompt. Hovering over the image shows a preview of it.

You can attach images in multiple ways:

  • Drag and drop images from your OS or from the Explorer view
  • Paste an image from your clipboard
  • Attach a screenshot of the VS Code window (select the paperclip 📎 button > Screenshot Window)

A warning is shown if the selected model currently does not have the capability to handle the file type. GPT 4o is the only model currently supporting images. You can provide JPEG/JPG, PNG, GIF, and WEBP image types. Claude 3.5 Sonnet and Gemini 2.0 Flash will soon support images as well.

Experimental Copilot status overview

We are experimenting with a new centralized Copilot status overview that provides a quick overview of your Copilot status and key editor settings. You can enable the Copilot status overview with the VS Code setting chat.experimental.statusIndicator.enabled. It includes the following changes:

  • Quota information if you are a Copilot Free user
  • Editor related settings such as Code Completions
  • Useful keyboard shortcuts to use other Copilot features

This Copilot status overview is accessible via the Copilot icon in the Status Bar.

Screenshot that shows the Copilot status overview in the Status Bar.

Experimental TypeScript context for inline completions

We are experimenting with enhanced context for inline completions and /fix commands for TypeScript files. The experiment is currently scoped to Insider releases and can be enabled with the chat.languageContext.typescript.enabled setting.

Custom instructions for pull request title and description

You can provide custom instructions for generating pull request title and description with the setting github.copilot.chat.pullRequestDescriptionGeneration.instructions. You can point the setting to a file in your workspace, or you can provide instructions inline in your settings. Get more details about customizing Copilot in VS Code.

The following sample shows how to provide a custom instruction inline in settings.

{
 "github.copilot.chat.pullRequestDescriptionGeneration.instructions": [
   {
     "text": "Prefix every PR title with an emoji."
   }
 ]
}

You need to have the GitHub Pull Requests extension installed to generate a title and description.

Accessibility

Copilot Edits accessibility

We made Copilot Edits more accessible.

  • Files with modifications and changed regions (insertions, modifications, and deletions) have audio signals.
  • The accessible diff viewer is now available for modified files. Just like in diff editors, select F7 to enable it.

activeEditorState window title variable

We have a new window.title variable, activeEditorState, to indicate editor information such as modified state, the number of problems, and when a file has pending Copilot Edits to screen reader users. When in Screen Reader Optimized mode, this is appended by default and can be disabled with accessibility.windowTitleOptimized:false.

See more

You can now attach images and work with them directly in Copilot Chat in VS Code or Visual Studio. Share screenshots of errors and Copilot will interpret the image and resolve the issue. Or share mockups of new designs, and Vision will help you bring them to life.

You can now attach images using several methods:

  • Drag and drop images from your OS or from the Explorer view
  • Paste an image from the clipboard
  • Attach a screenshot of the editor window (in VS Code, select Attach > Screenshot Window)

Currently, the supported image types are JPEG/JPG, PNG, GIF, and WEBP.
You can use the vision capability with the GPT-4o model. Make sure you have the Copilot ‘Editor Preview Features’ policy enabled to get access.

This feature was previously available to VS Code Insiders and Visual Studio Preview users, and is now available to users of the stable editor versions in public preview.

Copilot Chat analyzing an image

To learn more, read the documentation about using Vision in Copilot Chat.

Please share your feedback in our community discussions.

See more

The GitHub Copilot plugin for JetBrains IDEs now includes the ability to view code references, designed to enhance your coding experience and improve productivity.

What’s new ✨

Viewing code references: When GitHub Copilot suggests code that matches public code, you will be notified of this match. Click “View matches” at the end of the response to access the reference information.

Benefits for developers ⚡️

Informed decisions with code suggestions: code referencing allows you to make more informed decisions about whether to use the suggested code.

Get involved 🛠

We encourage you to try out the latest version of the GitHub Copilot plugin and share your feedback. Your input is invaluable in helping us refine and improve the product.

Join the discussion 🚀

Connect with us and other developers in the GitHub Community to share your experiences, ask questions, and provide feedback.

See more

Copilot Workspace header

This week, we’ve added an indicator for your daily or hourly quotas, allocated a section of the dashboard to issues assigned to you, and introduced several UX improvements and bug fixes to ensure a smooth development process with Copilot Workspace.

Showing approaching quota limits

As you’re nearing your quota limit, Copilot Workspace will now display a counter so you can keep track of how many tokens you have left, and prioritize your usage accordingly.

quota limits

Issues listed in homepage dashboard

Your recently assigned issues are now listed in the homepage dashboard, providing another jumping point to start developing from!

recently assigned issues in homepage dashboard

UX improvements and bug fixes

  • The session list has been refactored to link directly to pull requests created from a session.
    Screenshot shows list of recent sessions with linked pull requests
  • The sessions index page now maintains scroll position when navigating across sessions.

  • When viewing a new file’s diff, the diff is now highlighted in green.

  • We’ve added an arrow to the current branch name, indicating that it is clickable.

  • You can now select file names in the Plan stage, allowing you to copy them as needed.

  • We fixed the “live preview” button on the run command.

  • We fixed the overlay of the file tree view on small viewports.

  • Light/dark mode is now fixed to match your settings theme.

  • Terminal commands are now correctly disposed of when exiting the command dialogue without saving.

  • There’s improved file name generation to remove excess backtick characters.

  • We fixed a bug where the plan was not being shown in mobile view.

  • The “Revise” button on file headers is now represented by a sparkling pencil icon instead of crosshairs.

  • There’s a new square circle button for stop buttons.

Providing feedback

Please give your feedback in our GitHub Discussion. We’d love to hear your thoughts!

See more

GitHub Copilot GPT-4.5

OpenAI’s latest model, GPT-4.5, is now available in GitHub Copilot Chat to Copilot Enterprise users. GPT-4.5 is a large language model designed with advanced capabilities in intuition, writing style, and broad knowledge. It performs effectively with creative prompts and provides reliable responses to obscure knowledge queries. GPT-4.5 will launch in Visual Studio Code and on github.com for Copilot Enterprise users with a limit of 10 requests every 12 hours per user. In the coming weeks, we’ll be scaling rate limits and extending support to Visual Studio and JetBrains.

GPT 4.5 in the VS Code Model Picker

As model releases have continued to accelerate, we’ve been thinking about how we can sustainably offer advanced AI models like GPT-4.5 to more GitHub users. This includes individual developers who want the most advanced capabilities from day one. Stay tuned for updates.

Enabling access

Copilot Enterprise administrators will need to enable access to GPT-4.5 via a new policy in Copilot settings. As an administrator, you can confirm availability by checking your individual Copilot settings and confirming the policy for GPT-4.5 is set to “enabled”. Once enabled, users will see GPT-4.5 in the Copilot Chat model selector in VS Code and on github.com.

See more

Copilot Code Review

No more wait: Copilot code review is here

Code review is one of the most critical parts of software development. Manual code reviews, while essential, can be time-consuming. Copilot code review helps you offload basic reviews to a Copilot agent that finds bugs, potential performance problems, and even suggests automatic fixes. This means you can start iterating on your code while you wait for a human review—helping you keep your code repositories more maintainable and focused on quality.

Today we’re excited to announce that all Copilot subscribers can now use Copilot code review, with a host of updates that we’ve made since its initial preview. To get started, sign up here.

What’s new

  • Review summary: Copilot gives you a detailed summary of the changes in a pull request—bridging the gap between automated feedback and human insight.
  • Smarter reviews: We’re leveraging the latest models to provide even more insightful and effective feedback, identifying potential issues you might otherwise miss.

Here’s Copilot code review in action, providing a summary of the pull request and some suggested improvements:

To request a code review from Copilot, you can set up automatic reviews in a repo through repository rules. Or, if you prefer, you can ask Copilot to review a pull request on demand.

You can also jump directly into the new Copilot Workspace experience from your pull request, to refine and validate Copilot’s suggestions. And don’t forget, you can always review your code in Visual Studio Code before you push it to GitHub, keeping your development process agile and efficient.

To learn more, check out the docs. We can’t wait for you to try out these improvements, and we’d love your feedback in this GitHub Community Discussion.

See more

GitHub Copilot now available for free in Windows Terminal Canary

GitHub Copilot on Windows Terminal Canary is now available for free! GitHub Copilot Free allows you to access 50 chat messages per month. If you reach your quota, you can upgrade on the web.

GitHub Copilot in Windows Terminal

You can access the power of GitHub Copilot to get command suggestions and explanations without leaving the terminal with Terminal Chat in Windows Terminal Canary. This is available for all Copilot customers.

Get started today

GitHub Copilot is available in Windows Terminal Canary. Consult the Terminal Chat documentation to learn how to connect Copilot and get started.

Share your feedback

We are dedicated to continuous improvement and innovation. Your feedback remains a crucial part of our development process.
Learn more about GitHub Copilot Free and share your feedback on Terminal Chat.

See more

A stylized image showing a 3D Copilot head looking at a user message that has been edited via the new editing functionality

The immersive mode of Copilot chat on GitHub now provides you with increased control and flexibility at every step.

What’s new:

  • ✏️ Edit your previously sent messages: Refine your messages whenever you need
  • 🔄 Reload responses: Get different variations instantly
  • 📝 Keep track of all iterations in one place: Easily view the history of your message edits and responses
  • 🔍 Compare different message and response pairs seamlessly: Switch between versions to see which one works best

Pro tips:

  • 🛠️ Fix typos or add context without starting over
  • 🧪 Experiment with different phrasings to get the best responses
  • 🔀 Toggle between versions to compare outcomes

These new refining capabilities make it easier to iterate with Copilot. Make adjustments and improvements while maintaining the full context of your conversations, helping you to stay focused when chatting with Copilot chat.

We hope these changes help keep you in the flow state when chatting with Copilot on GitHub.com 💫.

💬 Let us know what you think using the in-product feedback option or pop it into the GitHub Community at any time.

See more

Claude 3.7 Sonnet is now available to all customers on paid Copilot plans. This new Sonnet model supports both thinking and non-thinking modes in Copilot. In initial testing, we’ve seen particularly strong improvements in agentic scenarios.

In GitHub’s internal evals, the model shows improvements over its predecessor in the ability to follow instructions, break down complex tasks, and build new UIs (human-reviewed).

Get started today!

Copilot Pro users

You can start using the new Claude 3.7 Sonnet model today via the model selector in Copilot in Visual Studio Code, Visual Studio, JetBrains IDEs, and immersive chat on GitHub.com.

Copilot Business or Enterprise users

Copilot Business and Enterprise organization administrators will need to grant access to Claude 3.7 Sonnet in Copilot via a new policy in Copilot settings. Once enabled, you will see the model selector in VS Code, Visual Studio, JetBrains IDes, and immersive chat on GitHub.com. You can confirm availability by checking individual Copilot settings and confirming the policy for Claude 3.7 Sonnet is set to enabled.

Claude 3.7 Sonnet is not currently available in Copilot Free, but you can access it by upgrading to a Copilot Pro plan.

Share your feedback

Join the community discussion to share feedback and tips.

For additional information, check out the docs on Claude Sonnet models in Copilot.

See more

Beginning February 2025, the beta Copilot /usage endpoints will be deprecated.
No new data will be inserted for retrieval via the beta /usage endpoints. This endpoint will be accessible through February, but the 28 day retention policy will remain.

Who’s impacted?

Enterprise organizations with Copilot licenses not yet migrated from the /usage metrics APIs are impacted by this deprecation.

What’s changing and why?

The deprecation of the beta /usage endpoints is a part of GitHub’s effort to deliver more powerful and flexible data offerings for enterprises, organizations, and teams. The new endpoints provide:

  • Visibility into the adoption and consumption of Copilot across various stages of dev lifecycle (from code suggestions to PR reviews), from the team to the enterprise level
  • Expanded scope of metrics, with the addition of GitHub.com Copilot Chat and Copilot for Pull Requests
  • Consistent terminology with the user management API
  • Better visibility into unique users at various drilldowns

Next steps

Ensure your organization is no longer consuming the now deprecated /usage endpoints in any jobs, workflows, and analytics tools.
As an alternative to the beta Copilot /usage endpoints, check out the PowerBI template and the Copilot /metrics endpoints.

Join the discussion in the GitHub community.

See more

Copilot Autofix helps you fix code scanning alerts and avoid introducing new security vulnerabilities by using large language models to suggest potential fixes.

We recently expanded the range of CodeQL security alerts where Copilot can suggest an autofix, covering a group that accounts for 29% of all CodeQL alerts. This expansion led to an 8% overall increase in alerts with an available autofix and a 270% increase in autofixes for this specific group of improved alerts. With more autofix suggestions, you can resolve security issues identified by CodeQL more easily—either by applying Copilot’s suggested fix directly or using it as a starting point for your own edits.

We made these improvements by analyzing our usage data to understand the most common types of alerts where Copilot was not suggesting fixes and then made a targeted effort to improve autofix for these alerts. Read more about the testing process that GitHub uses to identify the quality of autofix suggestions.

We continuously evaluate the performance of CodeQL and Copilot Autofix, so look for more improvements in the future.

See more