Visual Studio Code 1.133 landed on August 12, 2026, and unlike some recent releases that packed in a dozen features across every surface, this one is tight and focused. Three headline improvements, all aimed at removing friction from agent-driven workflows:

  1. You can now switch between Anthropic and Copilot model providers within a single Claude session
  2. The Agents window can be opened without forcing a GitHub sign-in
  3. Local HTML files auto-reload in the integrated browser

A fourth enhancement — sticky scroll for chat prompts — makes long conversations easier to navigate. Together, these changes reflect VS Code’s evolving understanding of how developers actually work with AI agents: not in single-provider silos, not behind unnecessary auth gates, and not with manual refresh steps for live previews.

Let’s break down what shipped.

Mix Anthropic and Copilot Models in Claude Sessions

This is the headline feature, and it addresses a real pain point for teams using Claude in VS Code.

Previously, a Claude agent session ran entirely through one of two paths: your GitHub Copilot subscription or Claude’s own API key configuration. Switching between them meant reconfiguring the agent host — not something you’d do mid-conversation. If you started a session on Copilot and realized you wanted to use a Claude model that’s only available via your Anthropic API key, you were out of luck until you reconfigured.

In 1.133, the model picker now displays both groups simultaneously:

  • Anthropic models — billed to your Anthropic API key
  • Copilot models — billed to your GitHub Copilot subscription

You select which one to use for the next turn. This means you can start a session with a Copilot-backed model for initial exploration, then switch to a Claude model via your API key for a specific task that benefits from Claude’s strengths — all without leaving the session or reconfiguring anything.

This is a meaningful shift in how multi-provider agent sessions work. In practice, different models excel at different things. Being able to pick the right tool for each turn — rather than committing to one provider for the entire session — is the kind of flexibility that makes a real difference in daily workflow. It also means teams with both Copilot subscriptions and Anthropic API keys can optimize cost and capability per turn.

The billing is transparent: models under the Anthropic group use your API key and are billed by Anthropic, while models under Copilot use your subscription. No surprises, no hidden routing — you see exactly which provider each model uses before you select it.

Agents Window Without GitHub Sign-in (Experimental)

Setting: chat.agentHost.allowSignedOutWhenUsable

The Agents window — VS Code’s dedicated surface for working with AI agents — has had a frustrating onboarding issue since launch: it opened with a GitHub sign-in prompt that you couldn’t dismiss. If your machine couldn’t reach github.com, or you simply didn’t need GitHub authentication, you were blocked.

This was particularly annoying for Claude users who already had an API key configured. They didn’t need GitHub at all, but the sign-in wall stood between them and the Agents window.

Enable this experimental setting and the Agents window opens without requiring GitHub sign-in. GitHub authentication becomes associated with individual agents or models rather than the window itself. So if you later decide to use a Copilot model, you authenticate at that point — not when you first open the window.

In this release, the no-sign-in behavior only supports Claude. Microsoft has stated that support for Copilot with your own model keys and Codex is planned for future releases. But for anyone who’s been locked out of the Agents window because of network restrictions, corporate policies, or simply not having a GitHub account, this is a meaningful unlock.

This also matters for air-gapped and restricted-network environments. If you’re working in an enterprise setting where GitHub isn’t accessible but you have an Anthropic API key, you can now use the Agents window directly. That’s a real workflow improvement for a segment of users who were previously excluded.

Auto-Reload HTML Files in the Integrated Browser

Setting: workbench.browser.autoReloadOnFileChange

When you have a local HTML file open in VS Code’s integrated browser, it now refreshes automatically when the file changes on disk. No manual refresh needed.

This sounds simple, but it eliminates a constant source of friction in agent-driven web development workflows:

  • An agent edits an HTML file and saves it
  • The browser tab updates immediately to show the change
  • You see the result without clicking refresh

You can toggle automatic reload for individual browser tabs, so you’re not forced into auto-reload if you want to keep a specific view static. The workbench.browser.autoReloadOnFileChange setting controls the default behavior.

Combined with the element-level commenting that shipped in 1.132, this creates a tight feedback loop: agent edits HTML → browser auto-refreshes → you review the rendered result → you annotate specific elements → agent iterates. The manual refresh step is gone, which means less context-switching and faster iteration cycles.

Sticky Scroll for Chat Prompts

Setting: chat.stickyScroll.enabled

If you’ve ever scrolled through a long agent conversation and lost track of which prompt produced the response you’re reading, this feature is for you.

When enabled, the prompt you’ve scrolled past stays pinned to the top of the chat — similar to how sticky scroll works in the editor. The pinned prompt shows its position in the conversation, and you can:

  • Click it to jump back to that prompt
  • Use previous and next buttons beside it to step through your prompts

This is one of those quality-of-life features that doesn’t make headlines but makes a daily difference. As agent conversations grow longer — especially with multi-turn refactoring sessions or extended debugging conversations — keeping track of the conversation structure becomes harder. Sticky scroll for prompts solves this elegantly without requiring you to scroll back up to find context.

Agent Host: Continuing Evolution

The agent host — VS Code’s dedicated process for running agent sessions based on the Agent Host Protocol (AHP) — continues to be the foundation for these improvements. The Copilot agent is powered by the Copilot SDK, which aligns its behavior with the Copilot CLI, the standalone GitHub Copilot app, and other Copilot products.

The ability to mix model providers within a Claude session is a direct consequence of the agent host architecture. Because sessions run in a dedicated process that’s connectable from multiple VS Code windows, the model picker can offer models from different providers without needing to tear down and recreate the session infrastructure. This is the kind of flexibility that the pre-agent-host architecture couldn’t support.

Community Contributions

This release had 13 community contributors:

  • Arthur Cnops (@accnops) contributed voice connection failure explanations and renamed Voice Mode voices to approved names — continuing their streak of voice-related improvements.
  • Simon Siefke (@SimonSiefke) fixed a memory leak in PerfModelContentProvider — extending their ongoing memory management work across recent releases.
  • Karthikeyan M (@karthikkatu) implemented the auto-reload integrated browser feature when a previewed local file changes — the core of this release’s headline browser improvement.
  • Benjamin Steenhoek (@bstee615) added optional rejected edit memory to the diffpatch prompt, helping agents learn from rejected suggestions.
  • Sanghyuk Jung (@benelog) fixed a duplicated word in a notebook setting description.
  • Tony (@Bosphoramus) fixed a top gap for Modern UI floating panels when the title bar is hidden.
  • Lori Fraleigh (@lfraleigh) added missing Azure SDK for Go modules to GoModulesToLookFor.
  • Mateus Abelli (@mateusabelli) updated Copilot extension links.
  • Ralph Feltis (@rfeltis) added an Agents window startup A/A experiment trigger.
  • Shehab Sherif (@ShehabSherif0) fixed operator precedence in timeline pageSize calculation.
  • Suraj Vaghela (@SVOG23) fixed stale relative paths in Claude chat sessions AGENTS.md documentation.
  • Bedirhan ÇETİN (@Xaena53) refactored find input toggle navigation to share it across views.
  • VS Code PR Bot (@vscodebot-pr) guarded against stale line numbers in test decorations.

Issue tracking contributors: John Murray (@gjsjohnmurray), RedCMD (@RedCMD), Andrii Dieiev (@IllusionMH), and Alberto Santini (@albertosantini).

The Bigger Picture

VS Code 1.133 is a smaller release than 1.132, but its improvements are precisely targeted at friction points that have been holding back agent workflows:

  1. Provider flexibility: Mixing Anthropic and Copilot models within a session eliminates the “commit to one provider” constraint. This is the most impactful change for teams that use both Claude and Copilot — and increasingly, teams do.

  2. Accessibility: Removing the GitHub sign-in gate from the Agents window opens VS Code’s agent capabilities to users who were previously blocked. For Claude users with API keys, the Agents window is now a direct entry point.

  3. Live preview: Auto-reloading HTML in the integrated browser removes the last manual step in the agent-edit-to-browser-preview loop. Combined with element-level commenting from 1.132, the browser is now a fully interactive development surface.

  4. Navigation: Sticky scroll for prompts addresses the practical reality that agent conversations get long, and losing your place in a conversation wastes time.

No new deprecations this release, which is a good sign — it means nothing is being removed while these improvements are being added. The agent host architecture is proving its value: the provider-mixing feature wouldn’t be possible without it, and more multi-provider capabilities are clearly on the roadmap.

If you’ve been waiting for the ability to use Claude without a GitHub account, or to switch models mid-session based on the task at hand, 1.133 delivers. Enable chat.agentHost.enabled and chat.agentHost.allowSignedOutWhenUsable to get the full experience.

VS Code 1.133 is rolling out gradually. Use Check for Updates in VS Code to get it immediately, or download the nightly Insiders build for early access to upcoming features.