On June 30, 2025, the Visual Studio Code team released the source code of the GitHub Copilot Chat extension under the MIT license, publishing it in the microsoft/vscode-copilot-chat repository on GitHub. The move completed the first milestone of a plan Microsoft announced on May 19, 2025, to make VS Code an open source AI editor. For the first time, developers could read exactly how Copilot's agent mode works, what context it sends to language models, which system prompts it uses and what telemetry it collects.

Key Facts#

  • Announcement: on May 19, 2025, the VS Code team said it would open source the GitHub Copilot Chat extension under the MIT license and then refactor its AI features into VS Code's core.
  • Delivery: the extension's source was published on June 30, 2025, in the microsoft/vscode-copilot-chat repository under the MIT license.
  • What became visible: the agent mode implementation, the context sent to language models, prompt engineering and system prompts, and the telemetry the extension captures.
  • What stayed closed at first: the original GitHub Copilot extension, which provided inline completions, remained closed source in June 2025.
  • Follow-up: VS Code 1.106, released on November 12, 2025, open sourced inline suggestions by merging them into the Copilot Chat repository and said the separate GitHub Copilot extension would be deprecated by early 2026.
  • Still a service: using Copilot continues to require a GitHub Copilot subscription, including the free tier.

What Happened#

The June milestone was mostly about transparency. The newly public repository contains the code behind Copilot Chat's conversational features and agent mode: how the extension gathers workspace context, how it builds prompts and calls tools, how it applies edits and how it recovers from errors. The VS Code team specifically highlighted that developers could now inspect the system prompts and the data the extension collects, two areas where transparency matters most for enterprise adoption.

The team was explicit about what was not yet included. Inline completions, the ghost-text suggestions most developers associate with Copilot, still came from the separate, closed-source GitHub Copilot extension. The plan was to move that functionality into the open source Copilot Chat extension over the following months, then refactor the relevant pieces into VS Code itself. The repository also shipped with contribution guidelines, and issues continued to be tracked in the main microsoft/vscode repository.

The May announcement had laid out five reasons for the change. According to the VS Code team, large language models had improved enough to reduce the need for secret prompting strategies, and the user interface patterns for AI interactions had converged across editors, so they were better refined in the open. Open code would also help extension authors build on VS Code's AI capabilities, let users see the data being collected, and let the community help find security issues faster, as it does in other open source projects. The team also promised to release its prompt test infrastructure so outside contributors could validate changes.

Background#

VS Code's core has been open source for years, but its AI features arrived through proprietary extensions. That split created friction. Extension authors building on VS Code's chat and language model APIs could not see how Microsoft's own reference implementation used them, and organizations evaluating Copilot had to rely on documentation to understand what left the developer's machine.

At the same time, the design space for AI in editors was maturing. Chat panels, inline chat, multi-file edits and autonomous agent loops had become common features across tools, which undercut the argument that prompt engineering was a competitive secret. Microsoft's bet was that openness would make VS Code the natural platform for AI tooling, with GitHub Copilot as the service that powers it.

Why It Matters for Developers#

For most developers, the practical effects are transparency and learning. The open repository is a detailed public example of a production coding agent, and it is useful even if you never contribute to it:

  • Study a real agent loop. If you are building agents in .NET with Microsoft.Extensions.AI or Agent Framework, the repository shows how a production system manages context windows, orchestrates tools, handles tool failures and asks for confirmation before risky actions. Our guide to AI agent architecture patterns covers the same concepts in C#.
  • Learn prompt design from a shipped product. System prompts from a shipped product are a better reference than toy examples. Compare them with the techniques in prompt engineering for .NET developers.
  • Audit data flows. Security and compliance teams can now verify what context the extension sends to models and what it logs, instead of relying only on policy documents. That supports the kind of review described in our responsible AI and LLM security guide.
  • Build better extensions. Extension authors can see how Microsoft's own code uses VS Code's chat, tool and language model APIs, which makes it easier to build compatible agents and tools.

There are limits worth keeping in mind. Open source code does not make the service free or self-hostable: requests still go through GitHub Copilot, and your organization's Copilot policies and plan still apply. The change also covered the VS Code extension only; other editor integrations were not part of the announcement. For a broader view of how to use Copilot and coding agents effectively in .NET projects, see AI-assisted .NET development with GitHub Copilot.

What's Next#

The roadmap Microsoft described in 2025 largely played out over the following year. VS Code 1.106, released on November 12, 2025, brought inline suggestions into the open source repository and announced that the separate GitHub Copilot extension would be deprecated by early 2026, leaving a single extension for suggestions, chat and agents. VS Code 1.116, released on April 15, 2026, made GitHub Copilot Chat a built-in extension, so new users no longer need to install anything to use chat, inline suggestions or agents, while a chat.disableAIFeatures setting turns AI features off for those who do not want them.

The standalone vscode-copilot-chat repository has since been archived, with development continuing in the main microsoft/vscode repository. That completes the arc from a closed add-on to a component of the open source editor itself. The open question now is less about code access and more about governance: how much community input will shape the agent features that now ship by default in one of the most widely used code editors.

Sources#