Visual Studio Code 1.132 landed on August 5, 2026, and it continues the steady march toward a more conversational, multimodal development environment. Where 1.131 introduced built-in dictation and the hybrid Markdown editor, 1.132 refines both — adding multilingual support to voice input, enabling Markdown diffs in the hybrid editor, and introducing a clever side chat mechanism that lets you ask questions without interrupting a running agent.

The standout feature, though, is element-level commenting in the integrated browser — the ability to point at a specific button, form field, or div on a web page and tell your agent exactly what’s wrong with it.

Let’s break down what shipped.

Element-Level Feedback in the Integrated Browser

This is the headline feature for anyone using VS Code’s integrated browser for agent-driven web development. When an agent builds or modifies a web app, you can now provide precise, element-level feedback without typing a general description and hoping the agent figures out which element you mean.

Here’s how it works:

  • Trigger element selection mode with a keyboard shortcut while in the integrated browser
  • Click one or more web page elements to select them
  • Add a comment to each selected element
  • Send all comments to the chat as structured agent feedback

This transforms the browser from a passive preview surface into an interactive annotation tool. Instead of “the submit button looks wrong,” you can click the actual button, write “this should be blue and say ‘Save’ instead of ‘Submit’,” and the agent gets exactly the context it needs.

Combined with the browser tools that went GA in 1.127 — where agents can open pages, take screenshots, click, type, and navigate — this closes the loop on agent-driven web development. The agent builds, you review in the browser, you annotate specific elements, and the agent iterates. No more guessing games about which element needs attention.

Multilingual Dictation with Nemotron 3.5

Setting: agents.voice.language

Built-in dictation debuted in 1.131 with an offline Nemotron model. In 1.132, it gets a significant upgrade: the default on-device model is now multilingual Nemotron 3.5, which supports multiple languages and can either follow your configured language preference or detect the language automatically.

Key improvements:

  • Automatic language selection: Dictation uses your system or browser locale when supported. Otherwise, the model detects the language from your speech.
  • Onboarding experience: First-time users get a guided setup with a live microphone waveform, device picker when multiple microphones are available, and links to dictation settings. Reopen it anytime via Voice Mode: Show Introduction in the Command Palette.
  • Customizable cleanup instructions: Run Voice: Configure Dictation Instructions to define project-specific terminology or formatting preferences. VS Code combines instructions from ~/.copilot/dictation.md and .github/dictation.md in trusted workspaces, layered on top of built-in cleanup rules that preserve meaning and prefer numerals where appropriate.
  • Foundry Local import: If network restrictions prevent the on-device model from downloading, the error notification includes an action to import an official Foundry Local model package from disk — important for air-gapped enterprise environments.
  • Download progress UX: During first-use download, the microphone button shows progress instead of dumping download text into the input field.

Audio stays on your device throughout. This is not a cloud-based transcription service — the model runs locally, which matters for privacy-conscious organizations and compliance-restricted environments.

Side Chats with /btw

This is one of those features that sounds simple but changes how you work with agents in practice.

When an agent is mid-turn — processing files, running subagents, executing commands — and you want to ask a quick question, you previously had two options: wait for the turn to finish, or interrupt it. Neither is ideal when you just want to ask “what does this function do?” or “why did you choose this approach?”

Now you can type /btw in the chat input to open a side chat that:

  • Shares the context and prompt cache of your primary chat, so it has full visibility into the current turn
  • Runs without interrupting the main agent turn
  • Lets you ask contextual questions about what the agent is doing right now

You can also select text in a chat response and ask contextual questions about that specific response. And for cross-chat context sharing, you can reference other chats by dragging chat tabs into the input box or typing #chat: followed by the chat title.

This is particularly useful when:

  • An agent is working on a large refactoring and you want to understand a specific decision without pausing the work
  • You’re reviewing agent output and want to ask “what if we did X instead?” as a parallel exploration
  • You need to verify the agent’s reasoning mid-turn rather than discovering an issue after the turn completes

Markdown Diffs in the Hybrid Markdown Editor (Experimental)

The hybrid Markdown editor introduced in 1.131 — which combines rendered Markdown with in-place editing and agent-actionable comments — now supports diffs.

When an agent modifies a Markdown file, the diff can open in the hybrid Markdown editor instead of a traditional text diff. You see:

  • Rendered Markdown with formatting intact
  • Gutter indicators highlighting added, changed, and deleted content
  • Full editability — the modified document remains editable while you review

Switch between text diff and the Markdown editor with diff annotations using the editor type dropdown. This means you can review documentation changes in the format they’ll actually be rendered in, rather than reading raw Markdown diffs and mentally rendering them.

For teams that use Markdown for documentation (which is most teams using VS Code), this makes agent-generated documentation changes substantially easier to review.

Agents Window: Track Session Activity from Chat Input

New live status pills appear above the chat input to help you follow what’s happening across an agent’s work:

  • Changes: Shows the number of changes. Click to view a live-updating multi-file diff for the current turn.
  • Previews: Access Markdown previews for files the agent creates or edits.
  • Subagents: Follow what a subagent is doing by opening its work in a separate chat.
  • Browsers: Follow along while the agent interacts with the integrated browser.

These pills update based on the session you’re currently viewing, so they’re always contextually relevant. This is a natural extension of the subagent visibility work from 1.131, bringing the same observability philosophy to all aspects of agent activity — not just subagents, but file changes, previews, and browser interactions too.

Switch Editor Types in the Agents Window

Setting: breadcrumbs.showEditorType

When multiple editor types are available (e.g., text editor, diff editor, hybrid Markdown editor), a new dropdown in the breadcrumb bar lets you switch between them directly — no more Reopen Editor With command. You can also set the default editor from the dropdown.

This is a small UX improvement that reduces friction when working with the increasingly diverse set of editor types in VS Code: text, diff, Markdown preview, hybrid Markdown, custom editors from extensions. The breadcrumb bar already provides file navigation, so the editor type dropdown is a natural placement.

Chat: Terminal Output Reflow

Expanded terminal output in chat now reflows to the available width when you resize the view. Previously, output used a fixed width that caused lines to wrap too early and left wasted space in wider views.

This applies to terminal output from both the Local agent harness and the Copilot harness running on the agent host. It’s a small thing, but if you’ve ever tried to read a wide terminal output that was wrapping at 80 characters in a 200-character-wide panel, you know why this matters.

Terminal: Shell-Aware Dictation Cleanup

Terminal dictation now applies shell-aware cleanup, so spoken commands preserve shell syntax. For example, saying “git commit dash m hello world” produces git commit -m "Hello World" instead of inserting the words literally.

This is the dictation equivalent of syntax-aware editing — the cleanup rules understand that you’re dictating to a terminal, not writing prose, and format accordingly. Paired with the multilingual dictation support, this means you can dictate terminal commands in your preferred language and get correctly formatted shell syntax out.

Proposed APIs: Custom Editor Priorities by Mode

The customEditors.priority proposed API lets extensions set different priorities for text editors and diff editors independently. An extension can use its custom editor by default for regular files while keeping the built-in editor as the default for diffs, or vice versa.

Existing single priority values continue to work for backward compatibility, and diff editors use explicit priority by default. Microsoft plans to promote this feature to stable in the next release, which will give extension authors finer-grained control over how their custom editors integrate with VS Code’s editing surface.

Deprecation: ChatAgentHostEnabled Policy

The ChatAgentHostEnabled policy has been removed. Administrators can no longer centrally disable the agent host through policy. Individual developers can still use the chat.agentHost.enabled setting to choose whether agents run in the separate agent host process.

This signals confidence in the agent host’s stability — Microsoft is comfortable enough with it to remove the administrative escape hatch. For organizations that were holding back on agent host adoption via policy, this is the signal to evaluate it directly through user-level settings instead.

Community Contributions

This release had strong community participation, with 27 contributors submitting PRs. Notable contributions:

  • Simon Siefke (@SimonSiefke) continued their memory leak cleanup streak from 1.131 with six more fixes: titleBarPart, notebook view model, decorationAddon._decorations, chatServiceImpl, settings-tree, and historyService. Simon is quietly doing the Lord’s work on VS Code’s memory management.
  • Arthur Cnops (@accnops) contributed voice question carousels and concurrent question queueing for the voice experience.
  • Mir (@mirimadahmed) added voice warm capture before hands-free playback and made the coding agent voice-aware for a better voice experience.
  • Rushil Patel (@rushil-b-patel) added a copy button to code blocks in Markdown preview.
  • Praneeth Kodumagulla (@praneethhere) added COPILOT_HOME support for Copilot CLI state.
  • Dan Brown (@jdanbrown) fixed terminal tab titles to show “~” instead of “$HOME” — a small UX touch that Unix users will appreciate.
  • Yogeshwaran C (@yogeshwaran-c) contributed a remarkable 8 PRs including JSDoc continuation patterns, npm completions, package.json catalog support, search details tooltip, serve-web proposed API, folding ranges, terminal exit code detection, and module script scoping.
  • Dmitry Savy (@dsavy4) fixed stableStringify treating shared references as circular and a BidirectionalMap stale reverse entry bug.
  • Tobias Hernstig (@thernstig) fixed forwarded ports status bar item to toggle the ports view.
  • zmr-233 fixed a :has() selector performance issue that triggered workbench-wide style invalidation.

Issue tracking contributors: John Murray (@gjsjohnmurray), Momin Ahmad (@MominRaza), Norbert Palinkas (@palinkasnorbert), Andrii Dieiev (@IllusionMH), and Ganlv (@ganlvtech).

The Bigger Picture

VS Code 1.132 has three clear themes:

  1. Closing the feedback loop with agents: Element-level browser commenting and live activity pills make it easier to guide agents precisely and observe their work in real time. The browser is no longer just a preview — it’s an interactive annotation surface.

  2. Voice goes multilingual: Nemotron 3.5 with automatic language detection, shell-aware cleanup, and customizable instructions makes dictation practical for non-English speakers and terminal workflows alike. The on-device model means there’s no privacy tradeoff.

  3. Non-interrupting interaction: Side chats via /btw recognize that working with agents is a conversation, not a batch job. Being able to ask contextual questions without stopping the work is a meaningful UX shift.

The agent host continues its progressive rollout, the hybrid Markdown editor is maturing with diff support, and the deprecation of the ChatAgentHostEnabled policy signals growing confidence in the infrastructure. If you haven’t enabled chat.agentHost.enabled yet, 1.132 is a good time to try it.

VS Code 1.132 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.