Changelog
Back to All

v5.34.0

7 days ago by Ian Zygmunt

New Features

Search Scopes & RAG+ System

  • Multi-provider search fusion with score evidence preservation through Reciprocal Rank Fusion (RRF)
  • Search permission system with 6-step decision tree for agent access control (SearchScopePermissionResolver)
  • Reranker catalog with 4 providers: Cohere, Voyage, OpenAI judge, and BGE local with cost budgeting
  • External search providers support for Elasticsearch, Typesense, Azure AI Search, and OpenSearch
  • SimpleVectorDatabase in-process driver for development and agent memory scenarios
  • Streaming search with concurrent provider emission for real-time results
  • Search observability via MJSearchExecutionLog with performance metrics and cost tracking

Keyset Pagination

  • Keyset (seek) pagination for large dataset processing via RunViewParams.AfterKey
  • O(log N) performance for background jobs and bulk operations regardless of data depth
  • Automatic optimization for vector sync and geocoding maintenance operations
  • Graceful fallback to offset pagination for composite primary key entities

Computed Columns Support

  • Enhanced entity metadata with EntityField.IsComputed flag for SQL Server computed columns and PostgreSQL generated columns
  • Improved CodeGen base view generation avoiding unnecessary view materialization
  • Self-referencing foreign key support with computed name fields

Component System Optimization

  • Massive performance improvement for component loading (12-20s reduced to ~70ms)
  • Server-side registry caching with hash-based 304 support
  • Lightweight component summaries loaded on demand in Component Studio
  • 163 MB bulk load removal from ComponentMetadataEngine

Improvements

AI Agent Enhancements

  • Pre-execution RAG integration with search scopes for context-aware responses
  • Search scope access control with AIAgent.SearchScopeAccess enum (None, All, Assigned)
  • Artifact tool improvements with SearchResultSetToolLibrary integration
  • Enhanced agent form with search scope configuration UI

Memory Management

  • Comprehensive memory leak fixes across 226+ packages
  • ResultHistory caps and proper cleanup in queue systems
  • Task store optimization in A2A Server with shutdown handling
  • LRU cache improvements for provider and issuer caches
  • Streaming reset fixes in BaseLLM implementations

Search Engine Hardening

  • Input sanitization removing SQL LIKE wildcards to prevent injection
  • Defensive fusion filtering non-finite scores and malformed data
  • Graceful error handling for null/undefined queries and provider failures
  • Edge case protection for streaming and concurrent operations

UI and Developer Experience

  • Custom SearchScope forms with fusion weights, live preview, and A/B testing
  • Knowledge Hub dashboard with search metrics visualization
  • Enhanced AI model browser and agent configuration interfaces
  • Improved Angular components with better loading states and error handling

Bug Fixes

  • Authentication shell component stability improvements
  • Golden layout manager enhancements for better window management
  • Vector database consistency fixes across Pinecone, Qdrant, and pgvector
  • Geocoding maintenance scheduling optimization (weekly vs daily)
  • Export service reliability improvements
  • Form validation edge cases in entity communications

Technical Improvements

  • Database migrations for both SQL Server and PostgreSQL with comprehensive schema updates
  • Class registration optimizations with tree-shaking prevention
  • TypeScript strict mode compliance across all packages
  • Test suite expansion with 237 unit tests and 17 end-to-end scenarios
  • Documentation updates including new guides for Search Scopes, Keyset Pagination, and Soft Deletes