> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/ai-workbench/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/ai-workbench/_mcp/server.

# What's New in AI Workbench?

## Overview


Read this page to get information about:

- New features across all components (Desktop App, CLI, Service, and Credential Manager).
- Latest documentation updates and new example projects.
- Recent improvements and enhancements.
- Links to detailed [release notes](/ai-workbench/release-notes).

Check this page after each release to get info on what's changed.


## Release Version @@wbRel_curr_ver@@

### Desktop App

The following features and improvements were added to the Desktop App:

**onStart.bash Script**

- A new `onStart.bash` script runs each time the project container starts, giving you a place to initialize the running container without rebuilding the image.
- Edit the script from **Environment > Scripts** in the Desktop App alongside `preBuild.bash` and `postBuild.bash`.
- See more information in the [onStart.bash Script guide](/ai-workbench/how-to/environments/onstart-script).

**Ubuntu 26.04**

- AI Workbench now supports Ubuntu 26.04 for Desktop App installations.
- See more information in the [support matrix](/ai-workbench/reference/support-matrix).

### CLI

The following features and improvements were added to the CLI:

- `nvwb edit script` now supports `onStart.bash`. Open and edit the startup script directly from the CLI. See the [onStart.bash Script guide](/ai-workbench/how-to/environments/onstart-script).

### Service

The following features and improvements were added to the Service:

- Added the `onStart.bash` startup script, which runs each time the project container starts. The script runs as the project user with temporary passwordless `sudo` and can access the mounted project files. The UI reports a failed or timed-out startup if the script does not complete successfully.
- Added Ubuntu 26.04 support to full local and remote installer workflows.
- AI Workbench now selects between the open and proprietary NVIDIA kernel module driver when installing a missing GPU driver on Linux. See [AI Workbench Support Matrix](/ai-workbench/reference/support-matrix).

### Credential Manager

- Improved reliability of NVIDIA Inference credential validation.

### Documentation

**New Pages**

- Added the [onStart.bash Script guide](/ai-workbench/how-to/environments/onstart-script) for running initialization commands each time the project container starts.
- Added a [WSL 2 mirrored networking setup guide](/ai-workbench/reference/windows-full-local-reference#use-mirrored-networking) for Windows full local installations that use NVIDIA Sync with mDNS discovery.
- Added a troubleshooting section for [NVIDIA Sync mDNS connections on Windows](/ai-workbench/reference/troubleshooting/windows#troubleshoot-nvidia-sync-mdns-connections).

**Updated**

- Updated the [support matrix](/ai-workbench/reference/support-matrix), install guides, and location reference to include Ubuntu 26.04.
- Added Linux GPU driver selection behavior and failure scenarios to the [support matrix](/ai-workbench/reference/support-matrix) and install guides.
- Updated [release notes](/ai-workbench/release-notes) for the @@wbRel_curr_ver@@ release.

## Related Topics

- [Latest Release Notes](/ai-workbench/release-notes)
- Explore the [NVIDIA Developer Forum](https://forums.developer.nvidia.com/c/ai-data-science/nvidia-ai-workbench/671)