You can now import and export evaluation data.

Gemini 2.0 models are scheduled for deprecation on June 1st. Users will be automatically upgraded to Gemini 2.5 models on or after May 29th if they haven't manually updated. This applies to "flash lite" and "flash" variants. Quotas will also be adjusted automatically.

Feature

The following new tools are available

• Google Maps tools
• Confluence tools
• Jira tools
• SharePoint tools

You can now define custom HTTP headers for MCP tools.

You can now update settings for fallback behavior (error handling) in agent application settings.

You can now use agents as tools.

Feature

You can now execute tools either synchronously or asynchronously.

Feature

The gemini-3.1-flash-live model is the new primary model ID for Audio-to-Audio (A2A), replacing the legacy gemini-3-flash-native-audio name to align to GDM naming convention. A dismissal banner will now appear in the console for affected projects to guide users through this change.

You can now create static and dynamic variables.

You can now create rich response widgets for rich, interactive UI elements in your user interface.

Users can now filter conversation history by source (Simulator, Live, Eval) directly in the console to improve auditing and debugging.

Feature

Added the ability to export specific application versions through the ExportAppRequest API.

Change

Upgraded the Evaluation UI with "Naive Replay" modes for Golden Results, custom expectation tagging, audio recording for golden user entries, and enhanced error insights.

Change

Improved speech endpointing and fault tolerance on background noises. Also implemented a mechanism to recover the conversation from false-positive barge-in.

Change

Improved stability and reduce latency for Python code execution.

Customer Experience Agent Studio is Generally Available

Customer Experience Agent Studio (CX Agent Studio) is now Generally Available (GA). This unified platform, powered by Gemini, simplifies the creation of high-quality customer experience agents. This launch includes agents-building-agents capabilities, featuring "Start with AI," which lets developers instantly generate a starter agent and tools by providing documents and high-level instructions. This launch also delivers multimodal excellence, combining high-fidelity native audio models for natural voice personas with an out-of-the-box configurable chat messenger that supports rich multimodal interactions and customizable widgets. CX Agent Studio provides a secure, enterprise-ready environment to rapidly build, improve, and monitor your AI workforce.

Capabilities:

• Start with AI: Accelerate development by using generative AI to build a starter agent and associated tools automatically. Simply provide your existing documentation and a set of instructions, and the platform handles the initial setup.
• Configurable chat messenger: Deploy a fully functional, web-based client "out of the box" that supports text, voice, and multimodal interactions.
• Rich content widgets: Customize the user experience with configurable widget templates (including product carousels, order trackers, and quick actions) to fit your brand needs.
• Easy deployment: Embed the messenger into your web or mobile application with a simple HTML snippet generated directly from the Deploy tab.
• Flexible authentication: Secure your agent with built-in support for both Token Brokers (service account-based) and OAuth 2.0 (user-based) authentication flows.
• Native audio models: Create natural, brand-specific voice personas with high-fidelity audio models (powered by Chirp 3) designed to handle complex conversations with ease.
• Enterprise connectors: Seamlessly integrate with essential business tools using pre-built connectors
• Enterprise-grade security: Ensure that your data remains secure with robust identity management and compliance standards.