Release Date: July 1, 2026 | Version: 1.127.0

VS Code 1.127 is here, and it’s a significant release for anyone working with AI agents in their editor. Browser tools for agents reach general availability, the Agents window gets a major organizational upgrade, and there’s a deprecation that might affect your local model workflow. Let’s break down what matters.

🌐 Browser Tools for Agents: Generally Available

This is the headline feature, and it’s been a long time coming.

After several milestones in preview, browser tools for agents are now generally available and enabled by default. Agents can open pages in the integrated browser, read content and console errors, take screenshots, select, type, and navigate — all without an external MCP server.

What does this actually mean? You can now ask an agent to build a web app, run it, open it in the browser, click through the UI, and validate that it works — all in a closed loop inside VS Code. No more switching to an external browser to check if your agent’s code actually renders correctly.

Enterprise controls: Administrators can disable browser tools entirely via the BrowserChatTools policy, or restrict which domains agent tools can reach using ChatAgentNetworkFilter with allow and deny lists. If you’re in a regulated environment, you have granular control over what your agents can access.

Setting: workbench.browser.enableChatTools (enabled by default)

🔒 Per-Site Browser Permissions

The integrated browser now supports per-site permissions — just like a real browser. When a page requests access to capabilities like camera, microphone, geolocation, clipboard, Bluetooth, USB, serial, HID, or sensors, VS Code prompts you to allow or deny.

This matters because agents are now interacting with the browser. If an agent opens a page that requests camera access, you get a prompt — the agent doesn’t silently grant permissions. Manage permissions for any site from the Site Permissions browser menu item.

📂 Agent Session Groups and Drag-and-Drop

If you’ve been running multiple agent sessions (and who hasn’t?), the sessions list in the Agents window was getting unwieldy. 1.127 fixes this with:

Custom session groups. Create your own groups to organize related sessions. Collapse group headers to tidy up the list. Each group has quick actions: start a new session directly in the group, or mark all sessions as done with one click.

Drag and drop everywhere. Reorder sessions by dragging up or down. Drag session groups and workspace headers to rearrange the list. Drag a session onto a group to add it. Drop a session onto the Pinned section to pin it. Select multiple sessions and move them as a block.

This sounds small, but if you’ve ever had 15 agent sessions open and couldn’t find the one you need, this is a quality-of-life improvement that actually changes how you work.

📢 Chat Input Banners: CI and PR Comments Without Leaving

When a coding agent session has an open pull request, the Agents window now displays banners directly above the chat input:

CI failures. When checks on the PR fail, a banner shows how many checks failed (e.g., “2 of 5 checks failed”) with two actions:

  • Fix Checks — starts an agent fix automatically
  • Reveal Checks — opens the failing checks in the Changes view

Pull request comments. When new review comments come in, a banner shows the comment count with actions:

  • Address Comments — hands the comments to the agent
  • Reveal Comments — opens them in the editor

This keeps you in the conversation instead of hunting through different views. The agent can act on CI failures and PR feedback right where you’re already working.

💰 Subagent Credits: See What Delegation Costs

When an agent delegates work to a subagent, you can now hover over the subagent section in the chat response to see the AI credits used. This is a direct response to the opacity problem — teams have been running multi-agent workflows without knowing which subagent is burning through their credit budget.

Combined with the session-level cost tracking introduced in 1.126, you now have visibility at both the session and subagent level.

💬 Multi-Chat Sessions: Close, Reopen, Fork

Multi-chat sessions (multiple chats within a single agent host Copilot session) got several improvements:

  • Close and reopen. The X on a chat tab hides it rather than discarding it. Bring it back from the Conversations dropdown, where each chat has a checkbox to show or hide.
  • Delete permanently. Use the tab context menu and select Delete Chat.
  • Progress across all chats. The session shows as in-progress whenever any chat is working. Each tab surfaces its own progress. The session header Changes pill reflects combined edits from every peer chat.
  • Fork into a peer chat. Forking a conversation creates a new peer chat in the same session instead of a new top-level session. The forked chat inherits the conversation up to the fork point and runs independently from its siblings.

🔧 Editor Gutter Feedback for Agent Changes

When reviewing an agent’s changes, you can now leave feedback directly from the editor gutter. Hover over a line to reveal an Add Feedback glyph, then select it to drop a comment on that line. This is faster than switching to a separate review view and makes directing an agent to a specific spot in the code much more precise.

🛡️ Terminal Command Sandboxing on macOS and Linux

Approving every agent-invoked terminal command is tedious. Starting with 1.127, VS Code is rolling out sandboxing for terminal commands on macOS and Linux. Commands run with:

  • Network access blocked
  • Filesystem access restricted

The agent only asks for approval when a command needs to elevate and run outside the sandbox. This means fewer interruptions for safe, routine commands — but you still maintain control over anything that could touch the network or write outside the workspace.

Turn it off via the Permissions drop-down if you need to.

🏢 File-Based Managed Copilot Settings for Enterprise

Administrators can now deliver managed GitHub Copilot settings from a JSON file on disk, complementing the native MDM channels and account-based enterprise settings introduced in 1.125.

This is specifically useful for machines that aren’t enrolled in device management — air-gapped systems, Linux dev boxes, or environments using configuration management tools like Ansible or Chef.

File locations:

  • macOS: /Library/Application Support/GitHubCopilot/managed-settings.json
  • Linux: /etc/githubcopilot/managed-settings.json
  • Windows: %ProgramFiles%\GitHubCopilot\managed-settings.json

The file is only honored when MDM or account-based enterprise settings are not present, making it a fallback for edge cases rather than a replacement for proper MDM.

⚠️ Built-In Ollama Provider Deprecated

If you’re using the built-in VS Code Ollama provider to run local models with Bring Your Own Key (BYOK), this affects you.

The built-in Ollama provider is now deprecated. Ollama has an official VS Code extension that is the recommended way to use local Ollama models in chat.

What to do: Install the official extension and remove the built-in provider to continue using your Ollama models without interruption. The extension approach lets Ollama ship updates faster than waiting for VS Code releases.

This is part of a broader pattern — VS Code is moving toward an extension-ecosystem model for model providers rather than building them into the core editor.

🧰 Other Notable Changes

Onboarding tours (Experimental). The Agents window now has guided walkthroughs to help new users get up to speed with agent capabilities. Still experimental, but a sign that Microsoft is thinking about onboarding for AI-native development.

Better PR titles and descriptions. The Create Pull Request button in the Agents window now uses session context to generate accurate titles and descriptions, instead of generic placeholders.

Responsive sessions sidebar (Experimental). On narrow windows, the Agents window can auto-hide the sessions sidebar when there isn’t enough room. Setting: sessions.layout.autoCollapseSessionsSidebar.

/troubleshoot command. Diagnose agent behavior issues by analyzing session logs. Works with agent host sessions, including local and remote. Type /troubleshoot #session and describe the issue.

💭 The Big Hat Take

1.127 is the release where VS Code’s AI agent story shifts from “building the foundation” to “making it production-ready.” The key themes:

  • Browser tools GA means the closed-loop build-test-fix workflow is real, not a preview experiment. This is the feature that makes agents genuinely useful for frontend and full-stack development.
  • Session organization (groups, drag-and-drop, multi-chat improvements) acknowledges that people are running many agents simultaneously and need to manage them like projects, not tabs.
  • Cost visibility at the subagent level addresses the #1 concern we hear from enterprise teams: “How much is this actually costing us?”
  • Terminal sandboxing on macOS and Linux reduces the approval fatigue that makes autonomous agent workflows impractical.
  • Ollama deprecation is the right call — the extension model is better for everyone — but if you’re using local models, make the switch now before the built-in provider stops working.

Bottom line: Update immediately. Browser tools GA alone is worth it, but the session management improvements make this a release that changes how you’ll work with agents day-to-day.

Already running 1.127? We’d love to hear how browser tools GA is working for you in practice — reach out and let’s compare notes.