March '25 Release Notes

New Features

• Semantic Layer: Advanced document analysis and processing to prepare complex PDFs as XML documents for LLMs. Vertesia creates a semantic layer describing PDF as XML by dynamically processings each page based on the nature of the content, preserving content hierarchy, mapping complex tables and charts, and describing images, without ever altering the fidelity of text. This feature enables the most advanced document preparation for LLMs to understand complex PDFs.
• New Image Generation Models: Added support for image generation and editing using Google Imagen 3 models. This enables flexibility to generate images using multiple providers and models.
• Model Tools: Added support for tools with Claude and Gemini models. This allows Claude and Gemini models to leverage external tools during generation, expanding their capabilities and use in agents that have access to tools.
• Async Tasks: New API endpoint to execute interactions as asynchronous tasks that run in the background, improving user experience for long-running tasks. This feature simplifies running interactions asynchronously without having to create new workflow. 

Improvements

• Top Navigation Bar: Enhanced the navigation experience with a new top navigation bar, including links to helpful resources and user information. A new user info dropdown provides a more structured and organized view of user and environment information, along with a quick way to change between light and dark modes. 
• Sidebar Enhancements: Refreshed the sidebar including organization and project switchers, layout, styling, and icons for a cleaner look. Desktop users can also now collapse and expand the sidebar on demand.
• Streamlined Project Configuration: Added options when creating a new project to streamline project configuration. "Project Description" provides additional context about the project to AI models and "Auto-configure the Project" automatically configures the project with shared environments, models, and sample prompts. 
• Improved User and Project Role Management: Refactored the Settings > Users page into separate pages for managing organization-level and project-level roles. This improves administration by providing dedicated spaces for managing users within an organization vs. project access. 
• Invitation Improvements: Added the ability to invite multiple users to an organization at the same time, streamlining the process of onboarding new team members. Also added a new section to display and manage pending invitations for an organization, allowing administrators to track and revoke pending invitations. 
• Stripe Billing: Implemented a new Billing page in the Settings for managing account billing through Stripe. This enables users to view their billing status and access the Stripe billing portal if applicable.
• Interaction Forking with Environment: Improved forking interactions by propagating the environment, model, and model options when the target project has access to the original environment.
• Enhanced Prompt Designer: Improved working with prompt templates by making it easier to add and edit properties, making the modal larger when editing prompt templates from an interaction, and fixing the header. 
• New Prompt Roles for Image Generation: Added support for new prompt roles using for image generation—Negative and Mask.
• Generated Image Filename: Improved the naming and labeling of generated images by uniquely identifying a file based on the interaction that generated it. 
• Upload Button: Added an explicit "Upload" button to the content object pages, for users who prefer using a file browser over a drop zone.
• Simplified Attention Mechanism: Simplified the attention mechanism in embedding generation to improve performance. 

Bug Fixes

• Organization and Project Switcher Refresh: Fixed a bug where the organization and project lists wouldn't refresh after accepting an invitation. The lists now update automatically after accepting an invitation.
• Incorrect Interaction Version: Fixed a bug where the wrong interaction version was displayed when publishing. The correct version number is now shown in the publish modal. 
• Infinite Table Reload: Fixed a bug where the content object table would infinitely reload if no objects were found. The table now displays an empty state message when no objects are available.
• Document Text Property Overwrite: Fixed a bug where the text property of content objects would be overwritten during document intake when a "Description" property existed. The text property now correctly reflects the extracted text or description.
• Model Options Clearing: Fixed an issue where model options were not cleared correctly when switching models in the interaction configuration. 
• Experimental Gemini Models: Removed experimental Gemini models from the Studio listing as Google restricts using them in production.