Testing the Open-Sourced GitHub Copilot Chat Extension for VS Code

Introduction

Microsoft has released the GitHub Copilot Chat extension for VS Code as open source software. It is available at github.com/microsoft/vscode-copilot-chat and it is licensed under MIT license. This is significant since the source code can now be viewed, built from scratch, and understood as to how Copilot works without using the paid service.

We needed to confirm whether the repo is an actual development toolset or only partially dumped. You can clone, compile, run the code on VS Code, and modify it. Can you plug your model into it? We did it all.

The Core Question

The primary question was rather straightforward: Does the open-source release give users the ability to build upon and enhance Copilot Chat, or is the code just a user interface on top of a closed API layer?

The pitch from Microsoft was transparency and community contribution, but our real question was: Can a company take the source code and adapt it for internal use?

The expectations before beginning:

1. Clone must run without needing any particular permissions.

2. It is based on a regular Node.js environment, nothing fancy.

3. It is built using TypeScript and VS Code extensions.

4. AI outputs may depend on Microsoft's servers nonetheless.

Repository Cloning and Setup

Step one was straightforward.

git clone https://github.com/microsoft/vscode-copilot-chat.git
cd vscode-copilot-chat
npm install

Dependencies installed clean. No auth walls, no license keys to chase.

Build and launch:

npm run compile
code --extensionDevelopmentPath=.

The README was clear. npm run compile finished without drama. Opening VS Code with --extensionDevelopmentPath=. loaded the extension in the Extension Development Host and the Copilot Chat panel showed up locally.

No surprises on setup. If you have Node.js and VS Code, you are in.

Code Structure and Modules

The extension is fully written in TypeScript. The code is modular; that means you can mess up with something without breaking everything.

src/chat/ Handles chat messages and generates responses. Here lies all the back-end parts.

src/panels/ All the frontend code related to chat interface buttons and prompts.

Copilot commands/src/commands/ Commands to be run via Copilot’s command palette

package.json Extension dependencies, activation events, and details about the extension.

The distinction between the panel (front-end) and chat (back-end functionality) becomes relevant once modifications are being done. Front-end modifications are independent; back-end modifications affect the API.

Modifications Attempted

Three types of experiments were conducted to determine how customizable things really are.

Customization of the UI

Modification of the code within src/panels/ modifications in button placements, prompts, etc. Recompilation through npm run compile. The changes instantly appeared on the VS Code interface.

Frontend Customization is a Snap. Doesn’t require any specific skills beyond React/TypeScript UI development.

Backend Logic Modification

In src/chat/ChatService.ts we considered the idea of redirecting all queries via our API (OpenAI or a local model). This approach seems feasible from an architectural perspective. However, the plug-in is hard-coded to rely on the Microsoft proprietary inference APIs. Authentication tokens are required for responses to be received.

It's all there. You cannot just take advantage of the built-in backend without any effort from your side.

Integration Testing with a Local Model

There has been work done to direct the message flow to a local architecture such as Code Llama from Ollama. The architecture will be able to accommodate this. The only drawback here is that you will have to handle all the requests and authorization mechanisms as developed by Microsoft.

Summary: Changes to the User Interface and behavioral level are fast. However, core AI features remain linked to the endpoints unless there is massive re-engineering.

Results and Findings

What You Can and Can’t Do

You can:

Modify the chat UI and prompts.

Expand upon commands found in the command palette.

Use the chat UI as a framework for your own chat application.

Learn about the integration of Microsoft’s AI Chat into the Visual Studio Code extension API.

You cannot (without putting in serious effort):

Use AI completions out-of-the-box without using the Microsoft Copilot service.

Replace the cloud-based model by tweaking just one configuration value.

Consider this a comprehensive, independent Copilot clone project.

It is licensed under MIT; therefore, you can fork it and customize it. However, the inference layer is missing from the repository.

Recommendations for Implementation

First make the UI changes if you are looking to score fast. The lowest-risk way to start is at src/panels/

Be sure to understand ChatService.ts before making any backend changes.

Iterate using the Extension Development Host instead of rebuilding the extension each time.

Do not assume open source implies self-hosted AI. Factor in engineering time for Ollama and OpenAI integration if necessary.

Maintain the repository. Microsoft is actively working on this, and your fork will drift.

Experiment with and without a Copilot license. Things behave differently if the commercial server is not present.

Conclusion

This open-source vscode-copilot-chat extension is real. Cloning works, building works, UI customization works. If you wish to explore or rebuild the Copilot Chat feature in Visual Studio Code, here you go.

The replacement of the AI brain itself is a totally different matter. The plugin is still dependent on Microsoft’s backend for its inference capabilities. You cannot use it without the aforementioned service or rewriting the entire network layer to make it fully functional; hence, you get an operational chat interface but no responses.

All the more reason for cloning, even though one half is open and the other is closed.

Technical Aspects

Repository:

github.com/microsoft/vscode-copilot-chat

License: MIT

Technology stack:

TS VS Code extension plugin

src/chat/ - messaging & responses

src/panels/ - chat widgets

src/commands/ - commands palette integration

Build steps:

npm install

npm run compile

code --extensionDevelopmentPath=.

Main Observations:

Project cloning and building - no issues

Modifying UI interface - easy

AI engine - proprietary, not provided

Replacing local AI model - possible, laborious

Reference:

• vscode-copilot-chat on GitHub