v0.5.0-beta - Advanced RAG Capabilities

🎉 isar_agent_memory v0.5.0-beta

Advanced RAG Capabilities Release

This is a major feature release that introduces 8 powerful new capabilities to significantly enhance the RAG (Retrieval-Augmented Generation) system.

---

🚀 New Features

1. Memory Consolidation

Automatically merge and deduplicate similar memories using LLM-powered clustering.

• findSimilarMemoryClusters() - Group related memories
• consolidateCluster() - Merge memories intelligently
• autoConsolidate() - Automatic memory optimization
• deduplicateMemories() - Remove redundant content

2. Embeddings Cache

LRU cache providing 10-100x performance improvement for repeated queries.

• Hit/miss rate tracking
• Automatic expiration
• Configurable size and TTL
• Access frequency analytics

3. Quality Metrics

Comprehensive RAG quality measurement and tracking.

• Relevance scoring
• Coverage analysis
• Latency tracking (avg, p95, p99)
• Query history and reports

4. Forgetting Mechanism

Intelligent memory cleanup based on importance and recency.

• Multi-factor importance scoring
• Age-based, LRU, and importance-based strategies
• Temporal decay with configurable half-life
• Memory protection for critical nodes

5. Dynamic Layer Creation

Automatic hierarchical memory organization.

• Adaptive clustering
• Layer optimization
• Graph structure analysis
• Automatic recommendations

6. Multi-Modal Support

Support for multiple data modalities beyond text.

• CLIP adapter (text + images)
• ImageBind adapter (all modalities)
• CodeBERT adapter (source code)
• Structured data processing
• Hybrid multi-modal adapter

7. Agent Memory Types

Cognitive memory systems inspired by human memory.

• Episodic: Events/experiences with temporal context
• Semantic: Facts and knowledge
• Procedural: Skills and procedures
• Working: Short-term memory with TTL
• Automatic episodic → semantic consolidation

8. Cross-Encoder Re-ranking

Advanced re-ranking for better search relevance.

• Cross-encoder models
• Hybrid reranker (multiple strategies)
• MMR (Maximal Marginal Relevance)
• Remote and local adapters

---

🛠️ Improvements

Code Quality

• ✅ Fixed 16 compilation errors across 7 files
• ✅ Resolved all analyzer warnings (0 issues)
• ✅ Clean dart analyze output
• ✅ Professional code formatting

API Fixes

• Corrected Isar collection access patterns
• Fixed type inference issues
• Added proper import statements
• Removed invalid @override annotations

Documentation

• Added comprehensive ADVANCED_FEATURES.md guide
• Usage examples for all features
• Best practices and performance tips

---

📚 Documentation

• CHANGELOG.md - Full changelog
• ADVANCED_FEATURES.md - Advanced features guide
• README.md - Getting started

---

⚠️ Breaking Changes

None. This release is fully backward compatible with v0.4.0.

---

📦 Installation

dependencies:
  isar_agent_memory: ^0.5.0-beta

---

🔗 Links

• Package: https://pub.dev/packages/isar_agent_memory
• Repository: https://github.com/iberi22/isar_agent_memory
• Issues: https://github.com/iberi22/isar_agent_memory/issues

---

💡 Notes

• Privacy features are still in development
• Some features require LLM integration via LLMAdapter interface
• Multi-modal features require additional model integrations
• All exported features are production-ready

---

Full Changelog: v0.4.0...v0.5.0-beta

v0.4.0 - HiRAG Phase 2, Cross-Device Sync & Re-ranking

🚀 Major Features

HiRAG Phase 2

Complete implementation of advanced hierarchical RAG capabilities:

• Added LLMAdapter interface for LLM integration
• Implemented �utoSummarizeLayer() for automatic layer summarization using LLMs
• Implemented multiHopSearch() for hierarchical context-aware retrieval
• Support for configurable prompt templates and search depth

Cross-Device Sync Backends

Real-time synchronization infrastructure:

• Added SyncBackend interface with pluggable backend architecture
• Implemented FirebaseSyncBackend for Firebase Realtime Database
• Implemented WebSocketSyncBackend for custom WebSocket servers
• Added CrossDeviceSyncManager for managing sync lifecycle
• Factory pattern for automatic backend selection
• Stream-based real-time change propagation

Advanced Re-ranking Strategies

Improve search relevance with multiple algorithms:

• Added ReRankingStrategy interface for extensible re-ranking
• Implemented BM25ReRanker for term frequency-based ranking
• Implemented MMRReRanker for Maximal Marginal Relevance (diversity + relevance)
• Implemented DiversityReRanker for maximizing result variety
• Implemented RecencyReRanker for time-based relevance
• Integrated re-ranking into semanticSearchWithReRanking() and hybridSearchWithReRanking()

🔧 Improvements

API Enhancements

• Extended MemoryGraph with semanticSearchWithReRanking() method
• Extended MemoryGraph with hybridSearchWithReRanking() method
• Added createdAt as optional parameter in MemoryNode constructor
• Improved query handling in semanticSearch() for layer filtering

Testing

• Added 7 new comprehensive test suites
• All core tests passing (13/13 ✅)

📦 Dependencies

• Added irebase_core and irebase_database for Firebase backend
• Added web_socket_channel for WebSocket backend
• Updated mockito to ^5.4.4 for improved testing
• Fixed: Added lutter_test to dev_dependencies

📖 Documentation

• Updated README.md with comprehensive documentation for all new features
• Added usage examples for HiRAG Phase 2, sync backends, and re-ranking
• Updated TASKS.md with completed sprint and new priorities

⚠️ Breaking Changes

None. All changes are backward compatible with v0.3.0.

---

See CHANGELOG.md for complete details.

v0.3.0: Sync Protocol and HiRAG

v0.3.0: Sync Protocol and HiRAG

Major Features

Sync Protocol: Implemented secure synchronization with Last-Write-Wins (LWW) conflict resolution using UUIDs.

• Added SyncManager class for encrypted export/import of memory graphs
• Added EncryptionService using AES-256-GCM for client-side encryption
• Extended MemoryNode and MemoryEdge with sync fields: uuid, modifiedAt, version, deviceId, isDeleted
• Documentation: Added doc/SYNC_PROTOCOL.md describing architecture and reconciliation strategy

HiRAG (Hierarchical RAG): Foundation for hierarchical knowledge management.

• Added HierarchicalMemoryGraph extension with createSummaryNode and getNodesByLayer methods
• Added layer field to MemoryNode for multi-level knowledge organization
• Support for summary and part-of relationships between nodes

Improvements

• Model Updates: Regenerated Isar and ObjectBox models to support new sync and HiRAG fields
• Testing: Added integration test for sync functionality
• Testing: Added unit tests for EncryptionService and SyncManager

Dependencies

• Added cryptography: ^2.9.0 for encryption support
• Added json_annotation: ^4.9.0 for JSON serialization

Links

• pub.dev package
• CHANGELOG
• Sync Protocol Documentation

v0.2.3

0.2.3 - 2024-11-24

• Fix: Critical bug in OnDeviceEmbeddingsAdapter - corrected outputs.values.first to outputs.first.
• Fix: Corrected invalid Mermaid diagram syntax in README (ASCII art → valid Mermaid).
• Fix: Moved http from dev_dependencies to dependencies for tool scripts.
• Improvement: Replaced forEach with for loops in resource cleanup (better Dart practice).
• CI/CD: Disabled automatic benchmark workflow trigger (manual only) and added Linux desktop support.
• Chore: Added test_resources/ and tool/ to .pubignore to reduce package size.

v0.2.2

See CHANGELOG.md for details.

v0.2.1

See CHANGELOG.md for details.

v0.2.0

See CHANGELOG.md for details.