Skip to content
  • There are no suggestions because the search field is empty.

August '25 Release Notes

Product release notes for August 2025

New Features

  • Specialized Workers With Activity Routing: We've implemented a new activity routing system that directs resource-intensive tasks, such as PDF processing and rendition generation, to specialized workers. This improves overall system performance and resource utilization.
  • Low-Priority Processing Queue: Users can now specify a `low_priority` processing option for content objects and asynchronous executions. This routes resource-intensive tasks, such as bulk imports, to a dedicated queue, ensuring that high-priority operations remain fast and responsive.
  • Update Collection Properties Tool: A new agent tool, `update_collection_properties`, has been added to allow agents to modify the metadata properties describing a collection.

Improvements

  • Automatic Plan Completion For Agents: Agents using plans will now automatically attempt to complete any remaining tasks, improving the reliability and completeness of multi-step operations.
  • Advanced Collections Search And Faceting: The collections API now supports complex search queries and faceting, allowing for more powerful and flexible filtering of collections in the UI and via the API.
  • Semantic DocPrep File Size Limit: The maximum file size for advanced PDF processing has been increased from 100MB to 200MB, allowing for the intake of larger documents.
  • Enhanced Virtual Load Balancer: The Virtual Load Balancer can use any model set during execution and provides more accurate run details. Fallback models are chosen based on their configured weights for better performance and reliability.
  • Retry Attempts For LLM Activities: The maximum number of retry attempts for certain long-running LLM activities, including document and embedding generation, has been increased to improve the success rate of these operations during long-running bulk imports.

Bug Fixes

  • Correct URL Generation When Replacing Files: Fixed a bug that caused a 404 error when creating a new version of a content object by replacing its file. 
  • Intake Workflow Triggering On Content Type Change: Fixed a regression where changing a content object's type did not trigger the necessary intake workflow. Properties are now correctly generated when changing the type.
  • Rendition Generation Error Handling: The error code for in-progress renditions has been changed from 500 to 412 (Precondition Failed), providing more accurate feedback to clients and reducing unnecessary error alerts.
  • Model Selection In Project And Interaction Settings: Fixed issues where the selected model would persist incorrectly when changing environments in projects and interactions.
  • Display Of "Model(s) Used" Metadata: Fixed a regression where the "Model(s) Used" field for content objects was not being populated. This metadata is now correctly captured and displayed.