Visual Studio Code 1.128 is out today, and it continues the rapid pace of AI-first editor improvements we’ve seen all year. This release focuses on three areas: making agent sessions more flexible and navigable, graduating multimodal support to general availability, and giving developers more control over their editor environment.
Here’s a technical breakdown of what shipped and why it matters for your workflow.
1. Copilot Vision Is Now Generally Available
What it is
Multimodal support in VS Code Chat is now generally available. You can attach images and PDFs to any chat prompt by pasting them in, dragging and dropping, or using the context menu. Agents can also read images via tool calls during a session.
Why it matters
This is the most broadly impactful change in the release. Copilot Vision works across VS Code, github.com, and the CLI, and it’s enabled by default for Free, Pro, Pro+, Business, and Enterprise plans. For enterprise teams, this means developers can now paste screenshots of error dialogs, architecture diagrams, UI mockups, or PDF specifications directly into chat and get contextual assistance without describing everything in text.
The practical use cases are significant: frontend developers can paste a design mock and ask the agent to implement it, DevOps engineers can share error screenshots for instant troubleshooting, and teams can attach PDF architecture documents as context for code generation tasks. This eliminates the friction of manually describing visual context and makes agent interactions significantly more efficient.
2. Multi-Chat Agent Sessions Now Support Claude
What it is
Claude agent-host sessions, powered by Anthropic’s Claude Agent SDK directly in VS Code, now support multiple chats within a single session. You can run several related conversation threads under one Claude session instead of spreading them across separate top-level sessions.
Why it matters
This builds on the multi-chat foundation introduced for Copilot sessions in recent releases and extends it to Claude. You can now compare approaches, branch from an earlier turn to explore alternatives, and run work in parallel — all within a single grouped session. Each chat maintains its own history, title, and model selection, and everything restores with the parent session after a restart.
For complex tasks, this is a workflow accelerator. You might have your main chat implementing a feature while a peer chat writes tests in parallel and a forked chat explores an alternative implementation. The chats stay grouped under the same Claude session, keeping your workspace organized.
3. Quick Chats Without a Workspace
What it is
You can now start a chat in the Agents window without selecting or opening a workspace first. These “quick chats” appear in the Chats section and are useful for questions that aren’t tied to a specific folder.
Why it matters
Not every interaction with Copilot needs to be a full project session. Quick questions about syntax, API references, or general coding concepts can now happen immediately without the overhead of workspace setup. Quick chats restore after reload and remain separate from your workspace sessions, so they don’t clutter your project-focused work.
This requires the agent host to be enabled via chat.agentHost.enabled. The Pinned and Chats groups stay visible when empty by default, and you can use sessions.list.showEmptyDefaultGroups to hide them until they contain sessions.
4. Read-Only Subagent Chats (Preview)
What it is
When an agent delegates work to subagents, their transcripts now appear as read-only peer chats in the Agents window. You can monitor each subagent’s progress without interrupting or steering it from the main conversation.
Why it matters
Subagent observability has been a recurring theme in VS Code’s agent evolution. With read-only subagent chats, you get a dedicated view into what each worker is doing — accessible from the Conversations menu, the running-subagents chip, or the inline subagent pill in the parent transcript. The subagent title is shown when available, and all mutating actions are omitted so the worker transcript stays view-only.
For teams managing complex multi-agent workflows, this provides the transparency needed to understand what’s happening behind the scenes without risk of accidentally derailing a subagent’s work.
5. Configurable Browser Tab Placement
What it is
A new setting, workbench.browser.newTabPlacement, lets you control where integrated browser tabs open:
- activeGroup (default): Browser tabs open in the active editor group
- sideGroup: Browser tabs open in a dedicated, locked group to the side
- window: Browser tabs open in a dedicated auxiliary window
Why it matters
The integrated browser has become a core part of the agent workflow — agents use it to validate their work, and developers use it to preview changes. As browser usage increases, tab clutter becomes a real problem. The ability to isolate browser tabs in a side group or separate window keeps your code editor clean while still providing quick access to browser output.
Pages opened from an existing tab (via Ctrl+click on a link) still open in the same group as the parent, and you can always reorganize tabs manually.
6. OS-Level Keyboard Shortcuts
What it is
VS Code can now contribute OS-level keyboard shortcuts that work even when VS Code isn’t the focused application. Add "systemWide": true to any keybinding in your keybindings.json to make it OS-level.
Why it matters
This is a subtle but powerful addition. You can now trigger VS Code commands — like opening the Agents window, starting a new chat, or running a specific task — from anywhere on your system without first switching to VS Code. For developers who live in multiple windows (terminal, browser, documentation, chat), this eliminates the context-switching overhead of finding and focusing VS Code before executing a command.
Example keybinding for macOS to focus the Agents window:
{
"key": "cmd+shift+a",
"command": "workbench.action.openAgentsWindow",
"systemWide": true
}
7. BYOK Model Enhancements
What it is
Two new settings improve the Bring Your Own Key (BYOK) experience:
chat.agentHost.byokModels.enabled(Experimental): Use BYOK models when running sessions on an agent host. Enable and restart the agent host process.chat.byokUtilityModelDefault: Configure the default utility model for BYOK sessions — choose the main agent model, a GitHub Copilot model, or none.
Additionally, you can now configure sampling parameters (temperature and top_p) per custom endpoint model via the modelOptions object in your model JSON configuration. Set a property to a number to override VS Code’s default, or set it to null to omit the parameter and use the model server’s default.
Why it matters
Organizations that use custom model providers — whether for compliance, cost control, or specialized use cases — get finer-grained control over how those models behave in VS Code. The ability to configure sampling parameters per model addresses a real pain point: some providers have strict requirements around temperature and top_p, and previously VS Code sent default values that could cause rejections or unexpected behavior.
8. Deep Links to Specific Chats
What it is
vscode:// deep links can now open a workspace and focus a specific chat identified by the link’s session query parameter. The “Open in VS Code” action in the Agents window uses this behavior.
Why it matters
Deep links make it trivial to return to a specific conversation — you no longer need to open the workspace first and then hunt for the right session in the list. This is particularly useful for sharing agent session context with teammates or for quickly resuming work after stepping away.
9. Enterprise: Managed Copilot Telemetry with OpenTelemetry
What it is
Organizations can now mandate where GitHub Copilot sends OpenTelemetry (OTel) data through managed configuration. The telemetry block in Copilot managed settings controls:
- The OTLP export endpoint and protocol
- The OTel service name and resource attributes
- Exporter headers (e.g., authentication tokens for the collector)
- Whether prompt and response content is captured
- Whether developers can change content capture settings
A managed value always wins, taking precedence over environment variables and user settings. This applies to both the Copilot Chat extension and the agent host process.
Why it matters
For regulated industries, telemetry governance is non-negotiable. This managed configuration ensures that observability data flows to an approved collector without requiring each developer to set OTEL_* environment variables — and it prevents individuals from accidentally (or intentionally) redirecting or disabling telemetry. The ability to control whether prompt and response content is captured gives security teams the granularity they need to comply with data retention policies.
Keyboard Shortcuts for Chats in the Agents Window
New keyboard shortcuts streamline chat navigation in multi-chat sessions:
- Create a chat: Quick create without menu navigation
- Reopen last closed chat: Recover from accidental closes
- Next/previous chat: Navigate between open chats
- Quick switch between open chats: Fast tab switching
- Close active chat tab: Clean up without reaching for the mouse
- Delete active non-main chat: Remove peer chats you no longer need
- Searchable chat picker: Find any open or closed chat by name
These shortcuts are scoped to the Agents window and fall back to session-level behavior when there’s no chat-specific action.
What This Means for Enterprise Teams
VS Code 1.128 continues the trajectory established over the past several releases: making AI agents more capable, more observable, and more manageable in enterprise environments. The GA of Copilot Vision is the standout — multimodal support unlocks workflows that simply weren’t possible before and is available to all Copilot subscribers by default.
For teams evaluating or actively using Claude alongside Copilot, the multi-chat support for Claude sessions brings parity with the Copilot agent experience and makes multi-model workflows significantly more practical.
The enterprise telemetry management via OpenTelemetry is worth particular attention if you’re in a regulated industry. Combined with the existing managed settings infrastructure, this gives IT administrators comprehensive control over Copilot’s observability pipeline.
Getting the Update
VS Code is rolling out gradually to all users. Use Check for Updates in VS Code to get version 1.128 immediately, or download the nightly Insiders build to try features as soon as they land.
For the full release notes, visit the official VS Code 1.128 update page.
Kevin Kaminski is the Principal Architect at Big Hat Group Inc., a Microsoft partner focused on Azure, AI agents, and enterprise cloud architecture. Follow him on X for daily insights on AI development tools and enterprise technology.