Changelog
Start typing to search…
Back to All

v2.113.2

4 days ago by Jordan Fanapour

IMPORTANT
To prevent potential regressions when deleting conversations, copy the latest MJ_BASE_BEFORE_SQL file from the MemberJunction Bootstrap v2.113.2 ZIP download before running updates.

New Features

Parallel ForEach Execution in AI Agents

Added support for parallel execution mode in ForEach loops with configurable concurrency (default: 10), providing 5–10x performance improvements for I/O-bound operations such as web scraping, document processing, and API calls.

AI Agent Iteration Support

Introduced ForEach and While loop capabilities in Flow Agents with comprehensive loop configuration options, including collection paths, item variables, index variables, and error handling.

AI Agent Memory Management System

Implemented a complete AI Agent Memory System with notes, examples, and semantic search capabilities. Agents can now learn from past conversations and automatically inject relevant context using configurable strategies (Relevant, Semantic, Recent).

AI Agent Examples System

Added the new AIAgentExample table and an example-based context injection framework to improve agent performance, consistency, and contextual understanding.

Enhanced AI Agent Notes

Introduced a multi-dimensional scoping system for agent notes with source tracking, priority, and status management, improving context organization and memory resolution.

Form Builder Actions Package

Integrated with Typeform, JotForm, SurveyMonkey, and Google Forms—enabling AI agents to create, update, and analyze form responses programmatically.

Memory Manager Agent

Added an automated scheduled job that analyzes highly rated conversations every 15 minutes to extract valuable notes and examples for future agent use.

Agent Context Injection

BaseAgent now automatically injects relevant notes and examples before execution when enabled, using intelligent prioritization and semantic matching.

Multi-dimensional Memory Scoping

Notes and examples can be scoped by Agent, User, and/or Company with an 8-level priority resolution system for precise context retrieval.

Planning Designer Agent

Introduced a specialized agent for architectural planning and design tasks within the Agent Manager ecosystem.

Agent Administrator Role

Added a dedicated administrator role for managing AI agents with proper permissions and access controls.

File Storage Provider Search

Added SupportsSearch capability to file storage providers for enhanced content discovery and indexing.

Conversation Artifacts Query

Introduced an optimized query system for retrieving conversation artifacts with advanced filtering and performance improvements.

Mermaid Diagram Creation

Added an action for generating Mermaid diagrams programmatically.

Get File Content Action

Introduced a new action for retrieving file contents with improved parameter handling.

Improvements

AI Engine and Embedding Performance

Optimized embedding generation to eliminate unnecessary regeneration, reducing processing time from ~3 seconds to under 1ms.
Added semantic search with in-memory embedding caching for faster and smarter memory retrieval.

Agent Memory Management Enhancements

Improved agent memory planning and lifecycle management for better long-term stability and context continuity.

Enhanced Sub-agent Communication

Added configurable message modes and message limits to prevent prompt context pollution and improve multi-agent clarity.

Improved Component Specifications

Enhanced component specifications with detailed usage guidance and enforced “spec as source of truth” consistency.

Database Performance Optimizations

Improved cascade delete handling for multiple result sets and optimized conversation tracking with selective record change disabling for speed and reliability.

Conversation Artifact Management

Replaced ArtifactID with ArtifactVersionID for improved version control and artifact tracking across conversations.

Skip Artifact Retrieval Optimization

Streamlined retrieval using a single optimized query, replacing four separate RunView calls for better performance.

Metadata Synchronization Improvements

Enhanced metadata sync with smart field detection, virtual property recognition, and improved validation systems.

Agent Creation Workflow

Agents created through Agent Manager now automatically assign the creator as the owner.

Azure Configuration Flexibility

Made MSGraph provider Azure configuration optional to prevent MJAPI startup failures in non-Azure environments.

Copy Button Functionality

Fixed copy button behavior to use the active tab content instead of always copying raw JSON.

Loading States and UI Feedback

Added loading indicators for chat interfaces and improved feedback for smoother user experience.

Database Schema Updates

Comprehensive updates to support new iteration, note management, and memory system features.

Bug Fixes

  • Storage Provider Deactivation: Properly deactivates unused storage providers to prevent configuration conflicts.
  • Agent Step Configuration: Fixed loop body type constraints and validation for ForEach and While operations.
  • Artifact Permission Handling: Resolved artifact permission issues and ensured correct UI access permissions.
  • Query Parameter Processing: Added missing Query Parameter Processor to SQLServerDataProvider exports.
  • CASCADE Delete Operations: Fixed multi-result-set handling to properly delete entities such as Conversations.
  • Sub-agent Message Pollution: Fixed issue where sub-agent completion messages polluted main prompt context.
  • Race Condition in AIEngine: Resolved race condition affecting the actions array in multi-threaded scenarios.
  • Duplicate Package Files: Removed duplicate MJCore files from MJGlobal package and restored proper CLI bumping.
  • Schedule Job Type Records: Fixed missing primary keys in scheduled job type records to ensure proper scheduling.
  • Form Builder Security: Removed insecure API token parameters from form builder actions to prevent token exposure.
  • Agent Final Step Configuration: Enhanced ForEach and While support in final step processing.
  • Conversation Detail Loading: Improved conversation detail loading reliability with proper conversationDetailID handling.

Summary

This release delivers major performance improvements, introduces a full AI agent memory system, expands iterative and contextual capabilities, and improves developer tools, security, and database performance—advancing the foundation for intelligent, scalable, and self-learning AI agent behavior.