Cody for Visual Studio: Accelerating Development with AI-Powered Assistance Inside the IDE

The Challenge

As LLMs reshape developer workflows, integrating them seamlessly into existing tools became the next frontier. Sourcegraph's mission was clear: empower developers using Visual Studio with Cody’s AI-driven capabilities without sacrificing performance, compatibility, or user experience.

However, Visual Studio posed unique technical constraints:

• The IDE is based on the .NET Framework, with limited support for modern web UI standards.
• Unlike VS Code and IntelliJ, Visual Studio lacked native support for the latest web rendering controls.
• GitHub Copilot had deep integration into Visual Studio using internal APIs, replicating or extending that functionality required working with unstable or undocumented features.
• Ensuring cross-platform consistency with other IDEs while dealing with Visual Studio’s legacy architecture presented a non-trivial challenge.
• Testing a hybrid application combining native .NET and web technologies required innovative approaches to UI automation.

Sourcegraph needed a partner with deep expertise in both modern web technologies and legacy .NET environments. They found that partner in VirtusLab.

The Solution

Working closely with Sourcegraph, VirtusLab engineered a robust and forward-compatible architecture that brings Cody’s LLM-powered capabilities into Visual Studio with precision and resilience. The key innovations included:

• WebView2 integration
  Leveraged Microsoft’s Chromium-based WebView2 control to embed a modern web chat interface directly in the Visual Studio IDE. This ensured full HTML5 compatibility while avoiding outdated controls like WebBrowser.
• Modular architecture with separation of concerns
   VirtusLab divided the extension into three cleanly separated layers:
  • Visual Studio integration layer: Built with C#/.NET, manages IDE UI and interactions.
  • Cody core services: Handles core logic and feature coordination within the extension.
  • Cody agent: A standalone Node.js/TypeScript process responsible for LLM communication and document context parsing. Communicates with the extension via JSON-RPC over stdio.
• Reusable agent architecture
   The Cody agent, shared across VS Code, IntelliJ IDEA, and Visual Studio, ensured consistent behavior and reduced duplication. This portable component managed all interactions with LLM providers like OpenAI, Claude, and Gemini Pro.
• Fail-safe modular loading
   To minimize the risk from evolving internal Visual Studio APIs (like the Microsoft.VisualStudio.Language.Proposals namespace), VirtusLab developed an innovative dual-loading mechanism:
  • A main extension based on AsyncPackage.
  • A separately loaded MEF component for autocomplete and auto-edit features.
• This design allows backward compatibility (down to VS 17.0) and mitigates the risk of API deprecations crashing the extension.
• Robust UI automation testing
   VirtusLab implemented a two-tiered UI testing strategy:
  • Used VsixTesting for native extension tests.
  • Integrated Playwright for end-to-end testing of the web-based chat component.

Daily GitHub Actions runs validate integration across new agent versions and include automated Visual Studio screenshots to aid debugging.

The Results

With VirtusLab’s support, Sourcegraph delivered a powerful and stable AI extension for Visual Studio in record time. Key achievements include:

• Successful release of Cody for Visual Studio Preview on August 23, 2024.
• Cross-IDE consistency, with Cody’s behavior unified across VS Code, IntelliJ IDEA, and Visual Studio.
• Strong market reception, with over 19,000 installs and a Top 10 ranking in Visual Studio's AI tools category.
• Future-proof design, ensuring resilience to changes in Visual Studio’s internal APIs.
• Fully automated test coverage, enabling rapid iteration and continuous quality assurance.
• Real-time LLM chat, code completion, and contextual actions, seamlessly integrated into the developer workflow.

Sourcegraph now offers developers using Visual Studio an AI assistant that is fast, helpful, and integrated, further solidifying its position as a leader in AI-driven developer tooling.