Visual Studio Code 1.135 (Insiders)

Follow us on LinkedIn, X, Bluesky | Follow Insiders Changelog on X or Bluesky
Last updated: August 19, 2026
Welcome to the 1.135 Insiders release of Visual Studio Code.
These release notes cover the Insiders build of VS Code and continue to evolve as new features are added.
You can still track our progress in the Commit log and our list of Closed issues.
Happy Coding!
August 18, 2026
-
Add support for keeping the configured Copilot agent sandbox policy applied across approval modes, and honor enterprise-managed sandbox settings in the permissions UI so security controls stay consistent. #330773: Update sandbox slider for copilot harness
-
Add support for using the mouse Back button to return from a stacked agent chat to the sessions list instead of navigating editor history. #331125: Chat: Use mouse Back to return to the agent sessions list
-
Add support for routing lightweight voice requests, such as direct answers and app actions, without turning them into coding tasks. Voice interactions can continue while an agent session is already running. #329490: Voice mode: distinguish direct answers and app actions from coding requests
-
Add support for muting the microphone and toggling live transcription from the voice UI, so you can pause speech capture or reduce transcript noise without ending the session. #331367: Add mute-mic button and transcription controls for ChatGPT voice mode
-
Add support for referring to known projects by display name, project URI, or multi-root workspace context when creating or listing agent sessions. Delegated work starts in the intended workspace without requiring an exact path. #330948: Improve workspace resolution for agent orchestration
-
Add support for OS notifications when agent sessions need input, finish, or fail. Select a notification to open the correct window and session, so you can track long-running work across VS Code windows. #317195: Notify when Copilot agent sessions need attention or finish
-
Add support for starting dictation from the editor and terminal context menus, making speech input easier to discover where you are already working. #331368: Add dictation entry point to editor and terminal context menus
August 17, 2026
- Add support for using the Copilot Agent Host with a bring-your-own-key (BYOK) model while signed out of GitHub. #331180: Test: No Auth scenarios with Copilot Agent Host Harness
We really appreciate people trying our new features as soon as they are ready, so check back here often and learn what's new.

Facts Only

* Visual Studio Code version 1.135 (Insiders) was released on August 18, 2026.
* Support was added for keeping the configured Copilot agent sandbox policy applied across approval modes, honoring enterprise-managed sandbox settings in the permissions UI.
* Support was added for using the mouse Back button to return from a stacked agent chat to the sessions list instead of editor history.
* Support was added for routing lightweight voice requests, such as direct answers and app actions, without turning them into coding tasks while an agent session is running.
* Support was added for muting the microphone and toggling live transcription from the voice UI during a voice session.
* Support was added for referring to known projects by display name, project URI, or multi-root workspace context when creating or listing agent sessions.
* Support was added for OS notifications indicating when agent sessions need input, finish, or fail.
* Support was added for starting dictation from the editor and terminal context menus.
* Support was added for using the Copilot Agent Host with a bring-your-own-key (BYOK) model while signed out of GitHub.

Executive Summary

Visual Studio Code version 1.135 (Insiders) was released on August 18, 2026. The release notes detail several new features focused primarily on improving the integration and functionality of Copilot Agent features and voice interactions within the editor environment. New capabilities include supporting enterprise-managed sandbox settings for Copilot agents to maintain consistent security controls across approval modes. Additionally, improvements were made to agent session management, allowing users to reference known projects by display name, URI, or workspace context when creating sessions, enabling delegated work to start in the intended workspace. Further enhancements address voice mode functionality, enabling routing lightweight voice requests like direct answers and app actions without converting them into coding tasks, and adding controls for muting microphones and toggling live transcription during agent sessions. Finally, dictation input has been added to context menus in both the editor and terminal, and OS notifications can be used to signal when agent sessions require input, completion, or failure.

Full Take

The release focuses heavily on deepening the integration layer between AI agent functionality, security policies, and multimodal input within the IDE environment. The pattern observed is an architectural shift where ambient, context-aware interaction—voice commands, system notifications, and contextual referencing for workspace management—is being formalized as first-class features of the development experience. This suggests a move toward cognitive sovereignty through seamless control over complex, distributed agent systems. The inclusion of specific controls for voice session management (muting/transcription) and security synchronization (sandbox policies) indicates that complexity is being managed not just via code logic, but through intuitive UI interactions intended to reduce cognitive load for the user managing AI workflows. The implication is that the friction point in adopting advanced agent features shifts from mere functional capability to ergonomic integration, where control over agent states and input modalities becomes as crucial as the output itself. What mechanisms are needed to ensure these contextual controls do not introduce new security ambiguities when integrating disparate enterprise settings? What is the long-term relationship between granular session management and user trust in autonomous workflows?

Sentinel — Human

Confidence

This text appears to be authentic release notes for a software update, exhibiting the formal, specific language expected of technical documentation rather than synthetic prose.

Signals Detected

Moderate sentence length variance; clear, direct feature listing characteristic of release notes.

High internal coherence; adheres strictly to the format and purpose of software release documentation.

Standard, itemized structure typical of changelogs; no overt matching of external argumentative templates.

Claims are highly specific (version numbers, ticket IDs) and grounded in a clearly defined technical context, suggesting direct source material.

Human Indicators

The presence of specific versioning, feature names (e.g., Copilot agent sandbox policy), and internal ticket numbers (#330773) strongly suggests a direct artifact from an internal development or release channel.