visualstudiomagazine.com via Reddit

VS Code 1.120 Agents Window Demands Paid Copilot Plan

microsoft agents coding tools coding-tools agents

Key insights

  • VS Code 1.120's Agents window still mandates a paid GitHub Copilot Business or Enterprise plan despite supporting local model backends like Ollama.
  • Agents gain terminal read/write access and live browser tab context, meaningfully expanding what automated coding sessions can interact with.
  • Microsoft's local-model framing covers inference only; authentication and orchestration remain cloud-dependent, requiring an active internet connection.

Why this matters

Developer tooling vendors are converging on a pattern where local inference is permitted but subscription gates remain in place, and VS Code 1.120 is the clearest public example yet of that architecture being shipped to millions of users. For founders building on top of VS Code or competing in the AI IDE space, this signals that Microsoft intends to keep Copilot billing central even as the inference layer commoditizes. For technical leaders evaluating air-gapped or cost-sensitive deployments, the distinction between local inference and local autonomy is now a procurement-level concern, not just a technical one.

Summary

VS Code 1.120 ships the Agents window to Stable preview, giving developers a dedicated workspace for AI agent sessions and MCP plugin configurations that sits apart from the main Copilot Chat panel. The feature supports Ollama and Foundry Local as model backends, which Microsoft framed as local-model capability. The catch: despite that framing, the Agents window still requires a live internet connection and an active GitHub Copilot Business or Enterprise subscription to function. There is no offline mode. The local models serve as inference backends, but orchestration and authentication still run through Microsoft's cloud infrastructure. Essentially: (Microsoft, GitHub) are expanding Copilot's surface area while keeping the billing relationship intact. - Agents now have access to open terminals, foreground terminal read/write, and live browser tab context shared on demand. - Ollama and Foundry Local support is real, but it does not remove the Copilot subscription dependency or the internet requirement. - Local-LLM communities reacted negatively, having interpreted the feature as a path to genuine air-gapped or subscription-free use. The episode clarifies that Microsoft's definition of "local model support" means local inference, not local autonomy.

Potential risks and opportunities

Risks

  • Enterprises that piloted Ollama or Foundry Local expecting offline-capable Copilot alternatives may face procurement conflicts when IT security policies prohibit the mandatory internet connection.
  • Open-source AI IDE competitors (Continue.dev, Cursor, Zed) could accelerate adoption among the local-LLM community if they ship genuine offline agent workflows before Microsoft removes the subscription requirement.
  • GitHub Copilot Business customers who interpret the Agents window as a paid-tier feature may escalate contract disputes if the local model framing is reviewed by procurement or legal teams as misleading.

Opportunities

  • Continue.dev and other open-source Copilot alternatives can directly target the local-LLM community's frustration by shipping a credentialing-free Agents-equivalent workspace in the next 60-90 days.
  • Ollama and LM Studio gain positioning leverage with enterprise buyers seeking IDE integration without cloud lock-in, and can use this episode to sharpen their pitch against Microsoft's constrained implementation.
  • Security-focused IDE vendors (JetBrains with its AI Assistant, or air-gap-specialized tooling from companies like Tabnine) can market their subscription and connectivity models as compliance-friendly alternatives to VS Code's architecture.

What we don't know yet

  • Whether Microsoft has any roadmap commitment to offline or subscription-free Agents window operation, or if the internet dependency is architectural and permanent.
  • Which specific MCP plugins are supported in the Stable preview and whether third-party MCP servers can run without routing telemetry through GitHub infrastructure.
  • Whether Foundry Local's integration differs from Ollama's in any meaningful way for enterprise air-gap scenarios, given Microsoft owns both the IDE and the Foundry product.

Originally reported by visualstudiomagazine.com

Read the original article →

Original headline: VS Code 1.120 Ships Agents Window to Stable Preview — Local Model Support Still Requires Live Internet and Paid Copilot Plan