Powered by

# Release Overview

This release elevates platform efficiency by introducing critical Stop and Auto-Resume controls for LLM Benchmarking to optimize resource usage. We have also deepened our integration ecosystem with a new GitHub Connector and advanced JIRA capabilities, while streamlining Automated Workflows through intelligent data parsing and multi-context input support. Additionally, we are strengthening governance in the Enterprise Knowledge Garden by locking core configurations post-publish to ensure content consistency. Finally, user productivity is enhanced in EnterpriseGPT with new conversation search capabilities and improved response formatting.

# What’s New?

# PF LLM Optimization Hub - Benchmarking

  • Introduces Stop Benchmark functionality to halt ongoing benchmark runs on demand
  • Enables benchmarking to continue running in the background even upon navigating away from the benchmarking screen

PF Tools

  • Introduces GitHub Connector with comprehensive repository, branch, issue, pull request, commit, and search operations
  • Expands Confluence Connector with comment and label management capabilities
  • Expands JIRA Connector with advanced issue management, sprint operations, and project administration features

PF Automated Workflow

  • Enables automatic data type detection and parsing for workflow inputs
  • Supports multi-context input mapping across workflow components

PF Enterprise Knowledge Garden

  • Locks core processing configurations post-publish to maintain content consistency

PF Enterprise Digital Experts - Expert Agent Studio

  • Disables file-based tools in agent configurations where file context is not supported
  • Updated Knowledge Garden Selection Interface

EnterpriseGPT

  • Introduces search functionality within conversation history
  • Enhances response copy functionality to preserve LLM-rendered formatting

# Detailed Notes

# PF LLM Optimization Hub - Benchmarking

Stop Benchmark Execution
Users can now halt ongoing benchmark runs using a dedicated Stop Benchmark button, providing immediate control over long-running or incorrect executions.

Key Features:

  • The stop button appears when a benchmark is in progress
  • Immediately halts backend processing for remaining rows
  • Updates benchmark state to “Stopped” with clear status indicators

Impact:

  • Eliminates waiting time for incorrect or unnecessary benchmark executions
  • Reduces wasted computational resources and LLM costs
  • Improves user control and flexibility during benchmarking workflows

Background Benchmark Execution
Users can now safely navigate away while a benchmark is running. The benchmarking process continues in the background, and the interface seamlessly reconnects to the active session to display ongoing progress or final results upon return.

Impact:

  • Allows users to multitask or close sessions without interrupting long-running benchmarks
  • Ensures execution state and results are instantly available when reopening the screen
  • Eliminates the need to keep the browser tab active for the duration of the test

# PF Tools

GitHub Connector
The GitHub Connector enables comprehensive repository management and automation directly within workflows and agents

Supported Operations:

  • Repository Operations: Search repositories
  • Branch Operations: List branches, create branches
  • Issue Operations: List issues, create or update issues, read issues, add issue comments
  • Pull Request Operations: Create pull requests, list pull requests, read pull requests, merge pull requests
  • Search Operations: Search code, search issues, search pull requests
  • Commit Operations: Get commit details, list commits

Impact:

  • Automates GitHub workflows without external scripting or manual intervention
  • Enables agents to interact with repositories, issues, and pull requests dynamically
  • Streamlines development operations with integrated version control management

Expanded Confluence Connector
The Confluence Connector now supports comment and label management operations for enhanced documentation workflows

New Operations:

  • Get page comments
  • Add comment to page
  • Get page labels
  • Add label to page

Impact:

  • Enables automated documentation feedback and tagging workflows
  • Allows agents to retrieve and add contextual information to Confluence pages
  • Simplifies collaboration by integrating documentation management into workflows

Expanded JIRA Connector
The JIRA Connector now includes advanced issue management, sprint operations, and project administration capabilities

New Operations:

  • Issue Management: Add comments, add worklog, transition issue status, link issues, link to epic, create remote links, remove issue links
  • Agile Operations: Get agile boards, get sprints from board, get board issues, get sprint issues, update sprint
  • Project Operations: Get all projects, get project issues, create project version, get project versions
  • User & Field Operations: Get user profile, get worklog records, search fields

Impact:

  • Significantly expands automation possibilities for project management workflows
  • Enables end-to-end issue lifecycle management within Purple Fabric
  • Supports complex agile workflows with sprint and board operations

# PF Automated Workflow

Automatic Data Type Detection and Parsing
Workflow inputs now automatically detect and parse values into the correct data type, eliminating manual type conversion

Parsing Rules:

  • Valid JSON objects or arrays → Object/Array
  • Boolean literals (true/false) → Boolean
  • Null or undefined literals → Null
  • Numbers (integer, float, scientific notation) → Number
  • All other values → String

Impact:

  • Eliminates manual data type conversion and reduces configuration errors
  • Simplifies workflow design with intelligent input handling
  • Improves reliability of data flow between workflow components

Multi-Context Input Support
Workflow components now support multiple outputs from different components combined with manual inputs in a single field

Key Features:

  • Reference multiple outputs from different components as inputs
  • Combine manual values with dynamic outputs
  • Preview resolved input values before execution

Impact:

  • Enables more dynamic and flexible workflow designs
  • Reduces need for intermediate transformation components
  • Simplifies complex data aggregation scenarios

# PF Enterprise Knowledge Garden

Locked Core Processing Configurations Post-Publish
Core processing configurations are now locked after an Enterprise Knowledge Garden is published, preventing changes that could impact content consistency.

Locked Settings:

  • Chunking strategy
  • Embedding configuration
  • Table extraction
  • Image extraction

Impact:

  • Maintains consistency of processed content across all published Knowledge Gardens
  • Prevents accidental configuration changes that could affect retrieval quality
  • Provides clear messaging when users attempt to modify locked settings

# PF Enterprise Digital Experts - Expert Agent Studio

Updated Knowledge Garden Selection Interface
The “Add Knowledge” interface in the Agent Designer now uses the updated side panel design for consistency across conversational and automation agents.

Impact:

  • Provides consistent user experience across all agent types
  • Streamlines knowledge selection with a modern, intuitive interface

File-Based Tool Restrictions in Agents
Tools that rely on File data types are now disabled in agent configurations, as agents do not currently support file context.

Disabled Tools:

  • Google Drive
  • Amazon S3
  • Custom Functions requiring File inputs

Enabled with Restrictions:

  • Slack (file attachments disabled)
  • Send Email (file attachments disabled)
  • Custom Functions with optional File parameters

Impact:

  • Prevents configuration errors by disabling incompatible tools
  • Provides clear error messaging when file operations are attempted
  • Improves user experience with proactive validation

# EnterpriseGPT

Conversation Search
Users can now search within past conversations to quickly locate previous queries, responses, or important insights.

Key Features:

  • Search covers user queries, agent responses, and file names
  • Results highlight matched messages in the conversation timeline
  • Clear search control to restore full conversation view

Impact:

  • Dramatically reduces time spent manually scrolling through conversation history
  • Improves productivity for users managing multiple long conversations
  • Enables quick retrieval of specific information from past interactions

Enhanced Copy Functionality
The copy functionality now preserves the formatting rendered by the LLM, maintaining structure and readability.

Impact:

  • Maintains formatting integrity when copying responses to external applications
  • Improves usability for users who need to share or document agent responses

# Platform Enhancements & Fixes

PF LLM Optimization Hub - Benchmarking

  • Fixed an issue where the benchmark execution state was not visible when navigating between screens
  • Enhanced metric calculation accuracy for benchmarks that are interrupted
  • Improved file processing reliability to handle corrupted or invalid file formats
  • Resolved an issue to ensure benchmark rows remain reusable after failed executions
  • Resolved model parameter compatibility errors across different LLM providers

PF Tools- Connectors

  • Fixed GitHub label field validation for List, Create, and Update Issue operations
  • Corrected the Jira Read operation to support retrieval of multiple issue keys
  • Fixed Jira description field formatting when using multiple input variables
  • Resolved execution errors in the Jira Transition Issue operation
  • Resolved execution errors in the Jira Get Issue Link Type operation
  • Corrected the Jira Link Issues operation to ensure the link type directionality is applied accurately
  • Fixed multi-row processing for the Confluence Connector during benchmarking
  • Ensured default fields pre-populate correctly during Jira Read operations

PF Automated Workflow - Platform

  • Enforced edit restrictions to prevent changes to published workflows in read-only mode
  • Corrected workflow component output caching behavior during dynamic parameter updates

PF Enterprise Knowledge Garden

  • Fixed the search result display to properly show the loading state during initial queries
  • Resolved an issue where metadata persisted after deletion in Chunk Editor and Manage Data

PF Governance

  • Corrected validation logic for custom date ranges in the Overview module

EnterpriseGPT

  • Enhanced concurrent chat session processing to ensure queries are handled independently
  • Improved the handling of file deletions while an upload is in progress
  • Resolved issues with CSV and Excel file processing during Skills execution

Platform Administration

  • Fixed file type filtering restrictions during workspace image uploads

Platform Infrastructure

  • Enhanced platform stability with improvements to the cloud-agnostic architecture