Copilot Extensions now in Public Beta

[LiteBrite82]

Copilot Extensions now in Public Beta

We’re thrilled to announce the public beta release of GitHub Copilot Extensions, available starting September 9th, 2024. Whether you’re a developer looking to leverage existing extensions or a builder excited to create new integrations, Copilot Extensions now offers a platform to make that happen. Copilot Extensions allow developers to build or install integrations into Copilot Chat on GitHub.com and all supported editors (Visual Studio and VS Code) to increase productivity by seamlessly integrating external tools and context into the developer workflow.

What's New in the Public Beta?

• Open Access: All Copilot users, organizations, and enterprises (on supported plans) can now install and use Copilot Extensions.
• Build your own extension: Any developer or organization can create Copilot extensions, including private and public extensions for organizations, and the option to publish your extension to the GitHub Marketplace
• Extension Toolkit: We've released a comprehensive set of resources for builders; with a learning pathway, docs, the Copilot-Extensions organization with code samples and an initial SDK, and a CLI debugging tool.

Get Started Today

Explore the GitHub Marketplace to try a Copilot Extension, or check out our Getting Started Guide to build your own.

Feedback

We’re eager to hear about your experience, as a user or builder, during pubic beta.

For general feedback or feature requests around using and building extensions, please post in the comments below.
For detailed builder feedback to the GitHub product team, fill out the Extension Builders Survey.
For user feedback to a specific extension publisher, please file an issue in the user-feedback repo with a label for the publisher's listing name attached.

[kyleli626]

Feature request -

Chat participants, and thus copilot (and copilot chat) extensions, don’t seem to work for inline-chat

• possible pr
• closed issue

For example, try @ workspace in an inline chat. This seems to limit ability of extensions, especially since (in my experience), I tend to use more inline chat vs chat view.

[sharonlo]

That's great feedback @kyleli626! definitely looking into how we support extensions to inline chat :)

[Tolga1452]

I have a question (for the future versions of this beta). Will GitHub send references (such as selected code) to the endpoint along with the history, just like how GitHub Copilot works? Because I need to make a VS Code extension to get the code references from the IDE, but if I do that, the extension adds another participant to the chat and probably I need to implement a separate system within the extension, which would differ from the one my server currently uses to handle requests from GitHub, especially because I get different data from the VS Code API.

I also wonder if other chat extension capabilities will be added. Because I would like to send custom buttons, follow-up messages etc. to the chat. And the most disappointing part is I expected to create commands for my Copilot Extension without having to build an IDE extension.

I would like to have these features because with the current usage of Copilot Extensions (I also took a look at the dreamcode), we can only provide a chat experience to the users, which is far away from helping to code.

[D1M1TR10S]

Hi @Tolga1452!

Will GitHub send references (such as selected code) to the endpoint...

Currently, "references" within the scope of extensibility are sent. We are going to support additional context, like selected code, within the next two months. This will apply specifically to Copilot Extensions built as GitHub Apps. For VS Code Chat Participants, selected code is available as context. You can refer to the VS Code documentation for details on VS Code Chat specific capabilities.

I also wonder if other chat extension capabilities will be added...?

Capabilities like custom buttons and follow-up messages are common requests that we're tracking. While these aren't on the immediate roadmap, we are considering these features for future updates.

We understand that a lot of builders want more capabilities beyond chat, and we’re actively exploring ways to support deeper integration as we continue building out the platform.

[Tolga1452]

Thanks! ❤️ I'm looking forward to the future updates

[dipuharidasan]

Thank you for the public beta - it has great potential! I was able to build a simple chat participant VS Code extension using this guide, which sends user queries to our internal LLM and posts the API responses back to the Copilot Chat window.

In this scenario, does Copilot read the data from our internal LLM that is posted back to the chat? If so, will this data be treated the same as the prompts and suggestions mentioned in the Trust Center? I am a GH Enterprise user.

[rajbos]

As the entire chat is send back and forth, GitHub will see the response you send back to the client. So yes.

Also be aware that the response is also send to Azure OpenAI for the content and safety filtering.

[dipuharidasan]

Thanks, Rob. I assume that the same would apply to any Copilot Chat responses, and the same data privacy policies would be applicable in this case as well.

[htekdev]

Are we able to have private extensions created only for our organization organization?

[D1M1TR10S]

Hey @htekdev. Yes! You can create private extensions at the organization level.

The process will be the same as GitHub Apps (here's the link to our docs on how to do that).

Currently, we don't support internal enterprise-level apps or extensions that can be shared across multiple organizations under the same enterprise. But this is on our roadmap and scheduled to release during Universe at the end of this month.

[isidorn]

Hi VS Code PM here 👋

Just wanted to add to the initial announcement. You can also build Copilot Extensions in VS Code. This is an interesting path if you already have a VS Code extension. More details about this can be found here
https://code.visualstudio.com/docs/copilot/copilot-extensibility-overview
https://code.visualstudio.com/api/extension-guides/chat

Thanks 🙏

[yne]

Hi VS Code extension dev here.

My current company only use on-premise LLM, and I was hopping to integrate them in VScode using this lm and chat APIs but I got a little confused using them, could you clarify if

• Is it possible to use the VScode lm and chat APIs without a Copilot subscription?
  • If that possible: which extension samples showcase it ? (I tried 3 samples, all required copilot)
  • If that's not possible: why are they part of the VScode API and not in a copilot-specific repo ?

While testing, I triggered some Vscode-native chat dialogs (ex: "workbench.action.openQuickChat") yet my test extension was not called after entering some prompt.

[isidorn]

Hi there 👋

Is it possible to use the VScode lm and chat APIs without a Copilot subscription?

Right now no. But these APIs are designed to be generic, and this might change in the future depending on user feedback and potential scenarios.

why are they part of the VScode API and not in a copilot-specific repo

Design decision we did that we believe makes it easy to extend Copilot on the VS Code side.

While testing, I triggered some Vscode-native chat dialogs (ex: "workbench.action.openQuickChat") yet my test extension was not called after entering some prompt.

I suggest to do these two tutorials to help get started
https://code.visualstudio.com/api/extension-guides/chat-tutorial
https://code.visualstudio.com/api/extension-guides/language-model-tutorial

[rfon6ngy]

Hi thanks for the clarification @isidorn 👍 !

I'll patiently hope for chat API to evolve from a paywalled model to a more open contribution-based model, where developers can register they custom models/backend (proposed.chatProvider seems to go in that direction).

Yes I've indeed read all those lm/chat tutorials but this diagram made me think that my extension could offer an alternative language model to the chat API

My current scenarios are

• Feeding corporate data (not only code, but specification, requirements...) to any external server is obviously forbidden.
• In-house formal languages trained LoRA are more accurate than generic LLM.

Hope this might help change toward the "open" solution :)