GitHub Copilot in Visual Studio — September update

GitHub rolled out its September 2026 update for GitHub Copilot in Visual Studio 2026, delivering automated pull request reviews inside chat, an overhaul of Bring Your Own Model capabilities, and automated remediation for vulnerable NuGet dependencies. According to GitHub Changelog's report, the release targets deeper workflow integration across the IDE by shifting autonomous coding tasks directly to dedicated Model Context Protocol servers. The rollout spans every subscription tier, covering Copilot Free, Student, Pro, Pro+, Max, Business, and Enterprise licenses across the Community, Professional, and Enterprise editions of the IDE.

This breakdown details the concrete architectural changes, provider requirements, and configuration steps for engineering teams evaluating the release. The changes are particularly relevant to .NET developers managing dependency security in Visual Studio 2026, engineering teams requiring private inference endpoints through custom model providers, and developers navigating pull request reviews across multiple hosting environments.

What GitHub Copilot in Visual Studio — September shipped

The core update introduces an autonomous Git agent inside GitHub Copilot Chat capable of examining pull requests without requiring developers to switch to an external browser tab. Developers open GitHub Copilot Chat, select the Git agent, and attach a pull request as context directly from the Git Repository window. The agent inspects the pull request data and produces a structured summary complete with navigable links pointing to underlying review comments and modified files.

The release also introduces direct vulnerability remediation directly from the Visual Studio 2026 Error List. When NuGet Audit triggers security warnings, Visual Studio now displays a dedicated fixer action directly alongside the warning item. Triggering this fixer automatically opens GitHub Copilot Chat in agent mode, which then diagnoses the vulnerable packages and applies corrective updates using an integrated NuGet MCP server.

What changed for builders in GitHub Copilot in Visual Studio — September

Builders gain an upgraded Bring Your Own Model preview that operates across the Community, Professional, and Enterprise editions of Visual Studio 2026, enabled by default. The preview accommodates custom deployments alongside standard GitHub Copilot models, allowing development teams to route inference through their trusted internal endpoints regardless of whether the developer is currently signed in to GitHub. The supported model provider ecosystem officially covers Microsoft Foundry, OpenAI, Anthropic, and local Ollama deployments.

Under the hood, Bring Your Own Model transitions to a new Agent (Preview) runtime built on top of a dedicated GitHub Copilot SDK-powered harness. This change restructures how external models interface with Visual Studio workspace context. The shift ensures models interact consistently with tool definitions and runtime context, providing a unified harness for both hosted commercial endpoints like Microsoft Foundry or OpenAI and local developer hardware running Ollama instances.

How to install or upgrade GitHub Copilot in Visual Studio — September

Developers can obtain the latest capabilities by downloading or updating to the current build of Visual Studio 2026, with bleeding-edge builds accessible through the Insiders channel. The features do not require separate command-line installations, but teams configuring local runtime endpoints like Ollama can initialize and verify their models prior to IDE registration using standard terminal commands.

ollama run llama3

Within the Visual Studio 2026 interface, developers access the model switcher inside the Copilot Chat window to assign their preferred default or pinned inference provider for active sessions.

/model custom-endpoint

To configure dependencies for the automated fixers, developers open the Copilot Chat tools menu and enable the required integration toggles. Activating pull request inspections requires connecting the Model Context Protocol servers for GitHub or Azure DevOps, while vulnerability patching requires signing in to GitHub Copilot and enabling the NuGet MCP server option from the tools menu.

Gotchas and compatibility in GitHub Copilot in Visual Studio — September

The migration involves an explicit breaking change for existing Bring Your Own Model setups. The previous Ask and Agent modes have completely dropped support for BYOM, meaning any custom model configurations must operate strictly within the new Agent (Preview) harness. Furthermore, any developer who previously configured local Ollama models in Visual Studio will find those entries disconnected, requiring them to manually re-add their Ollama endpoints following the installation of the update.

Feature access also introduces specific service dependencies across external integrations. The Git agent cannot explore pull requests until developers connect their Git provider MCP servers, supporting either GitHub or Azure DevOps repositories. Similarly, while the core Bring Your Own Model feature functions without signing in to GitHub, the Error List fixer for NuGet vulnerabilities explicitly requires an active GitHub Copilot sign-in alongside an enabled NuGet MCP server in the Copilot Chat tools menu.

What to watch after GitHub Copilot in Visual Studio — September

Following this release, teams should monitor future capabilities arriving through the Visual Studio Insiders channel, where preview updates land prior to general availability. Upcoming releases are expected to expand the reach of the GitHub Copilot SDK-powered harness, deepening integration across enterprise source control environments and broadening protocol support for custom tooling.

Engineering leads should also track roadmap disclosures and release note updates published through the official Visual Studio blog. As the Agent (Preview) harness matures, teams using Microsoft Foundry, Anthropic, and OpenAI endpoints will likely see refined governance options and extended agent capabilities across testing and project management lifecycles.