Current Release Notes#
This document contains the release notes for NVIDIA AI Workbench.
Releases are identified by the pattern year.month.build-number.
Changes are identified for the following components:
AI Workbench Component |
Description |
|---|---|
Desktop App |
The AI Workbench Desktop Application |
CLI |
The AI Workbench CLI |
Service |
The AI Workbench Service |
Credential Manager |
The AI Workbench Credential Manager |
You can find the release notes for past versions of AI Workbench in Version History of Past Releases.
Release 2026.06.5#
Date: July 2026
AI Workbench Component Versions:
AI Workbench Component |
Version Number |
|---|---|
Desktop App |
|
CLI |
|
Service |
|
Credential Manager |
|
New Features and Improvements#
CLI#
The CLI has the following new features:
onStart.bash Script
The
nvwb edit scriptcommand now supportsonStart.bash. Use it to open and edit the startup script that runs each time the project container starts. See Use the onStart.bash Script.
Desktop App#
The AI Workbench Desktop App has the following new features:
Ubuntu 26.04
AI Workbench now supports Ubuntu 26.04 for Desktop App installations.
onStart.bash Script
The
onStart.bashstartup script is now visible in Environment > Scripts alongsidepreBuild.bashandpostBuild.bash.
Service#
The AI Workbench Service has the following new features:
onStart.bash Script
A new
onStart.bashscript runs each time the project container starts. Use it to initialize the running container after AI Workbench mounts the project. The script runs as the project user with temporary passwordlesssudoaccess. See Use the onStart.bash Script.AI Workbench monitors the startup script and base-image entrypoint and reports a failed or timed-out startup in the UI.
Existing projects may require one container rebuild to migrate to the updated container entrypoint before
onStart.bashtakes effect.
Ubuntu 26.04
Ubuntu 26.04 is now supported for full local and remote installations.
Linux GPU Driver Installation
On Linux, AI Workbench now selects a driver that uses either the open or proprietary NVIDIA kernel modules, depending on the installed GPU. See AI Workbench Support Matrix.
Credential Manager#
The AI Workbench Credential Manager has the following new feature:
Improved reliability of NVIDIA Inference credential validation.
Bug Fixes#
CLI
The following bug fixes were added to the CLI:
Opening a project in Cursor from AI Workbench now enforces
--classicmode.Fixed non-interactive installs on Ubuntu hosts with cross-device mount points.
Improved GPU detection on Brev instances.
Desktop App
The following bug fixes were added to the Desktop App:
On Linux, OAuth integrations (such as Brev and GitHub) now show their connected or error status immediately after you finish authorizing in the browser.
Fixed a bug where Brev locations could not be closed if the Brev instance was deleted.
Clicking Build Container no longer shows Build in progress while waiting for confirmation that files have been reviewed.
Remote support bundles now correctly contain
remote-ai-workbench-support-bundle.zip.Fixed a bug where the installer incorrectly skipped the WSL enablement step.
Windows full local installs now work when WSL 2 uses mirrored networking.
The
.projectdirectory is now visible in the file browser. You can view its contents and editspec.yaml, but you cannot upload, rename, delete, or duplicate other files within it.If NVIDIA Endpoints is disconnected, Build Assistant now shows Enable Build Assistant instead of allowing you to start the assistant from a failed build.
Clicking Learn More from an error dialog now opens the docs to the correct error code section.
Fixed the What’s this link for self-hosted GitLab integrations.
Added Linux app window keyboard shortcuts.
Service
The following bug fixes were added to the AI Workbench Service:
On Windows, AI Workbench now verifies that Docker Desktop preserves the required WSL configuration and that Docker is usable before setup continues, preventing installation from succeeding with an unavailable container runtime.
Fixed a macOS installer failure when detecting Xcode Command Line Tools on macOS 26.x.
GPUs reserved for a project container are now released if the container fails to start because of a base environment or startup script error.
Fixed an intermittent error when stopping Compose services.
AI Workbench now installs a driver using the open NVIDIA kernel modules when required for newer GPUs.
Credential Manager
The following bug fixes were added to the AI Workbench Credential Manager:
Fixed the redirect URI for the Brev integration.
Past Version Release Notes#
For more information about AI Workbench past releases, see: