Powered by

# Release Overview

This release introduces Enterprise GPT, a unified AI assistant for secure, complex reasoning, alongside significant enhancements to authentication management, benchmarking usability, and data export capabilities, empowering users with streamlined credential workflows, intelligent export controls, and contextual guidance for faster decision-making.

# Introducing EnterpriseGPT: Your Everyday Work Productivity Assistant

EnterpriseGPT seamlessly integrates agent skills, Knowledge Gardens, and powerful built-in tools to serve as a single, unified AI assistant for complex reasoning and insight generation.

Key Features:

  • Enterprise Resources: Unifies Knowledge Gardens and skills built from Agent and Digital Experts into a single interface
  • Powerful Built-in Capabilities: Access real-time web search, visualize data with customizable plots, and generate professional graphs directly within the platform
  • CSV and Excel Support: Enables on-demand file uploading with newly added support for CSV and Excel formats

Impact:

  • Accelerates time-to-answer by resolving multi-source queries (EKG, Skills, and Web) in a single interaction rather than sequential tool usage
  • Expands analytical depth allowing users to query and visualize complex datasets (Excel/CSV) instantly

# PF Enterprise Digital Experts - Tools & Authentication

Simplified Tool Authentication During Consumption
Users consuming workflows and agents can now define credentials upfront through an intuitive authentication flow, ensuring secure access before execution.

Key Features:

  • Clear authentication prompts within conversational agents via tool pill icons
  • The governance screen highlights tools requiring authentication with 'Authentication Required' status

Impact:

  • Eliminates execution failures due to missing credentials
  • Streamlines user experience with clear authentication guidance
  • Enables credential reuse across multiple consumption scenarios

# PF LLM Optimization Hub - Benchmarking

Flexible Metric Selection Without LLM Re-runs
Users can now select or deselect specific metrics in benchmarking, with scores updating without triggering additional LLM runs, reducing both cost and latency.

Key Features:

  • Metrics can be checked/unchecked to get an updated score by running metrics only
  • Row-level and summary-level scores updates without model execution
  • Selective row execution for targeted benchmarking runs across models

Impact:

  • Significantly reduces operational costs by eliminating unnecessary LLM calls
  • Accelerates iteration cycles with quick metric updates
  • Provides flexibility to experiment with different metric combinations efficiently

Default Export Controls for Summary Results
The export functionality now provides a default field selection for summary results ensuring only relevant data based on their selected result type.

Key Features:

  • Summary Results export automatically restricts selection to only Metrics
  • Input, Expected Output, and LLM Output options are disabled for Summary exports
  • Inline helper text clarifies that Summary reports include only aggregated metrics

Impact:

  • Prevents export errors by enforcing valid data combinations
  • Reduces confusion with clear guidance on what data is included in each export type
  • Streamlines exports by ensuring clean, relevant and selective data exports

#

# Enhancements and Bug Fixes

PF Enterprise Digital Experts - Expert Agent Studio

Resolved Benchmarking Display Issues
Fixed multiple issues affecting the benchmarking experience for conversation and automation agents, including UI display inconsistencies, file upload functionality in conversation benchmarks, and removal of unintended seed text from custom metrics output.

Resolved ReAct Agent Sub-Agent Visibility
Fixed an issue where agents did not display configured sub-agents in the Agent Designer interface, though they remained functional in the Prompt Preview. Sub-agents are now fully visible in the Agent Designer.

Corrected Prime Agent Tab Navigation
Resolved a display issue where tool details incorrectly appeared when switching from the Tools tab to the Sub-Agents tab in Prime agents, ensuring accurate content display during navigation.

Fixed Watermark Alignment in Published Agents
Corrected watermark positioning inconsistencies that occurred when editing published automation agents, maintaining consistent visual presentation.

Agent Designer Notifications for Component Removal
Users will now receive notifications when Knowledge Gardens, Tools, or Sub-agents are removed while building an Agent

# PF Enterprise Knowledge Garden

Fixed File Embedding Status Issues
Resolved an issue where Knowledge Garden file uploads became stuck in “Embedding” status, preventing files from reaching “Ready” state and becoming available for agent consumption.

Corrected Search Functionality for Deleted Files
Fixed a bug where deleted files continued to appear in Knowledge Garden search results, ensuring search accurately reflects current knowledge garden content.

Corrected File Import UI Alignment
Fixed misalignment of the on-demand upload icon on the Knowledge Garden import screen for improved visual consistency.

Improved Knowledge Garden Removal Notifications
Corrected notification messaging when Knowledge Gardens are removed from agents. The system now displays “Knowledge Garden Removed” instead of the generic “Tool Removed” message, providing clearer context for users.

Resolved Invalid Search Display Issue
Fixed an issue where the Import page was incorrectly displayed when performing searches with no matching results in Knowledge Garden. The system now properly shows “No results found” message.

# PF LLM Optimization Hub - Benchmarking

Resolved Intermittent Benchmark Processing Failures
Fixed intermittent “failed to process the request” errors occurring during conversation benchmarking with various LLM models, ensuring reliable and consistent benchmark execution.

Enhanced LLM Model Compatibility Handling
Implemented intelligent error handling for unsupported model parameters across OpenAI, Azure OpenAI, and AWS Bedrock providers, particularly when using advanced output features. Users now receive clear, actionable error messages when models don’t support specific parameters.

Resolved Temperature Parameter Compatibility
Fixed API failures when attempting to use non-default temperature values with models that only support default settings (e.g., GPT-5), preventing configuration-related errors.

Resolved File Deletion for Large Files
Fixed an issue where the delete icon was not displaying for benchmark files larger than 20MB, enabling users to properly manage all uploaded files regardless of size.

Resolved Contextual Accuracy Metric Failures
Fixed frequent failures in contextual accuracy metric calculations during benchmark execution, ensuring reliable evaluation of agent responses against expected context.

# Tools

Standardized Tool Naming Across Platform
Unified tool naming conventions across all platform sections including Tools, Instructions, Examples, and Traces for improved consistency, model accuracy, and simplified debugging.

Resolved Tool Version Display
Fixed an issue where tool version information was not visible on the Agents Tools screen when multiple versions existed, enabling clear version identification and selection.

Corrected Account Creation Error Messaging
Fixed duplicate error messages and improved validation feedback when creating new tool accounts in Flow Designer, providing clear, single error messages for validation issues.

Resolved Matplotlib Library Support
Fixed permission errors preventing Matplotlib plotting functionality in Custom Function V2, restoring data visualization capabilities and maintaining compatibility with popular Python data science libraries.

# PF Governance - Traces and Observability

Resolved RAG Source Reference Navigation
Fixed abnormal page refresh behavior and disappearing chunk point references when navigating between source file reference pages in conversational RAG agents, ensuring smooth navigation and persistent reference visibility.

Fixed Cache Behavior and Trace Logging
Corrected inconsistencies where cache was serving responses for on-demand files without proper indication in trace logs, ensuring accurate reflection of cache usage for debugging and performance analysis.

Resolved Thought Process Visibility in Traces
Fixed an issue where agent thoughts were not displaying in trace logs, restoring full visibility into agent reasoning and decision-making processes for debugging and analysis.

# Flow Designer

Resolved Component Duplication Issue
Fixed a bug where all Flow Designer components appeared as duplicates in the component palette, eliminating confusion and improving workflow creation efficiency.

Corrected Test and Debug UI Display
Fixed overlapping dropdown icons in the Test and Debug screen’s Output Field section (JSON and Pretty views), ensuring all UI controls are fully visible and accessible.

Resolved Tenant-Specific Access Issues
Fixed “Error 403” issues preventing Flow Designer and Custom Tool creation in specific tenant environments, restoring full platform functionality for all users.

Expanded File Size Support
Increased file size support from 20 MB to 50 MB for the S3 component within Flow Designer, accommodating larger data files and documents in automated workflows.

Resolved Agent Profile Description Display
Fixed an issue where profile descriptions were not displaying in the Agent Profile screen after updating or adding them in Edit Profile, ensuring complete profile information visibility.

# Platform Updates

Corrected Privacy Policy Link
Updated the Privacy Policy link on the login page to direct users to the official Purple Fabric privacy policy page, ensuring users access accurate and relevant privacy information.

Resolved User Login Issues
Fixed authentication flow issues that prevented new users from successfully logging in, where the system redirected back to the username page after password entry, ensuring seamless onboarding experience.

Improved Partner Policy User Experience
Refined the user interface for GenAI Partner policy users with the following enhancements:

  • Removed Help documents option from left menu for partner policy users
  • Eliminated empty workspace dropdown display
  • Fixed logout functionality to prevent unauthorized re-entry via browser back button
  • Corrected chat text box display when chat history window is opened

Resolved Automation Agent Table Output
Fixed errors occurring when attempting to add headers to table outputs generated by Automation Agents, enabling proper formatting of structured data outputs.

# Infrastructure Enhancements

Optimized Kubernetes Health Probe Configuration
Enhanced Kubernetes probe configurations for AI services (Query Engine, Model Hub, MCP) to prevent unnecessary pod restarts during high load conditions, improving platform stability during peak usage periods.

Stabilized MCP Gateway Dependencies
Pinned MCP Gateway requirements to stable versions to prevent unknown package-level issues, ensuring consistent and predictable gateway behavior.