Image: Microsoft / code.visualstudio.com VS Code 1.127: Multi-Chat Sessions, Terminal Sandboxing, and a Cost Meter for Subagent Work
VS Code 1.127 ships multi-chat support inside agent sessions, macOS terminal sandboxing, per-site browser permissions, and a hover showing how many AI credits your subagents burned.
VS Code 1.127 shipped July 1. It’s a dense update focused on making agent sessions more manageable when you’re running several things at once, plus a few additions to the security controls around what agents can touch.
Multiple chats in one session
You can now run several chat threads inside a single agent session. The agent panel tracks all of them under one progress bar, and you can fork any conversation into a peer chat to explore a different direction without losing the original. That’s a meaningful quality-of-life change if you find yourself constantly reopening sessions to try variations.
The session header now has Workspace and Changes pills that stay consistent regardless of what’s happening in the chat, making it easier to orient yourself mid-session.
Terminal sandboxing on macOS and Linux
Commands run from the agent terminal now use restricted network and filesystem access on macOS and Linux. This reduces how many approval prompts appear during a session, since the sandbox limits the blast radius without requiring you to approve every individual command.
There’s a dedicated toggle in the Default Approvals mode settings, so you can adjust the behavior per project.
Browser permissions, per site
Browser tools are now generally available (they were in preview). The new addition in 1.127 is per-site permissions: pages can now request access to camera, location, microphone, sensors, and the clipboard, and VS Code prompts you for each site individually rather than granting blanket access.
Subagent credit hover
When Copilot delegates work to a subagent and you hover over that section of the response, you’ll see how many AI credits that subagent consumed. This is the first place inside VS Code where you can see a cost breakdown at a per-step level, not just a session total. For teams watching spend closely, this closes a real visibility gap.
Agent session organization
The Agents window now supports custom groups. You can collapse headers, drag sessions between groups, and reorder things manually. It’s a small UX change but useful once you’re managing several long-running sessions.
/troubleshoot command
A new /troubleshoot command is available in agent sessions. It generates a diagnostic report of what the agent has been doing and where it got stuck — meant to help you figure out why a session went sideways without having to scroll back through the whole log.
PR generation using session context
The Create Pull Request button now reads from the agent session to write the PR title and description, rather than just using the diff. In practice this should produce more accurate PR descriptions since the session has more context than the raw changes.
Full release notes are at code.visualstudio.com/updates/v1_127.