**Source URL:** https://general.veevavault.dev/rn/26r2/ai

# 26R2 Vault AI Releases

An "AI release" is a release that contains features for Vault AI that may include new AI agents, enhancements, model updates, or minor adjustments. We apply these changes to General Release Vaults, and we communicate these changes on the day of the release.

AI releases are named using the format `AI.YYWW`, where `YY` represents the two-digit calendar year and WW represents the calendar week of the deployment. For example, `AI.2642` refers to the AI release deployed in week 42 (mid-October) of 2026.

## AI.2642

**Build Number**: 10946

**Release Date**: October 16, 2026

<ReleaseNote id="OOU0000000SD028" lr="AI.2642" app_family="platform" version="ai-26r2">
### Vault CLI

Vault CLI is an official, product-supported command line interface (CLI) for Veeva Vault. Vault CLI manages configuration, data, files, and instances. As a CLI, it has no GUI and is designed for power users, developers, scripting, and automation.

Vault CLI runs as **Headless**: Manual updates for tools, scripts, and power-users.

Users can authenticate through the browser or with an API access token. Browser authentication uses Vault as an OAuth server, just like Vault Mobile. It does not require API permission but customers can opt out.

Vault CLI releases on the [AI Release](https://rn.veevavault.help/en/gr/about-vault-releases/#vault-ai-releases) schedule, and works on Windows, Mac and Linux.
</ReleaseNote>

<ReleaseNote id="OOU0000000SK022" lr="AI.2642" app_family="platform" version="ai-26r2">
### Query Agent Enhancements: API/MCP

Query Agent and Vault AI metadata enhancements expand the Query Agent's capabilities by enabling Vault API and MCP access, improving data display, enriching metadata, and optimizing token efficiency.

The benefits include:

*   Enables Query Agent via Vault API and MCP (non-interactive mode) for programmatic access
*   Delivers user-friendly, localized data display through automated VQL post-processing
*   Reduces token consumption and latency via tool response history optimization
*   Enriches agent context with object types, document lifecycles, and a new knowledge base object
</ReleaseNote>

<ReleaseNote id="OOU0000000QG106" lr="AI.2642" app_family="platform" version="ai-26r2">
### Tool Call Invoking Agent Actions

Vault now supports tool calling an agent action, enabling AI agents to invoke other agent actions (sub-agents) as tools during LLM-driven orchestration. This capability supports both a programmatic Vault Java SDK approach for maximum flexibility and a zero-code configurable tool type for simple scenarios.

There are two ways to achieve this:

*   **Programmatic (Java SDK)**: Developers implement `AiToolTypeRuntimeHandler` to invoke one or more agent actions asynchronously within a tool call, with full pre/post-processing control.
*   **Configurable Agent Action Tool Type (UI/MDL)**: Admins configure a tool by selecting an agent and agent action, without the need for code. This supports custom objects and documents.
</ReleaseNote>