Changelog¶

All notable changes to GraphMem are documented here.

[1.8.3] - 2024-12-05¶

Added¶

• Comprehensive Agent Building Guide (docs/AGENT_GUIDE.md)
• Distributed Infrastructure Guide (docs/DISTRIBUTED_INFRASTRUCTURE.md)
• Better document validation error messages
• Shows type and preview of invalid documents

Fixed¶

• Fixed batch ingestion validation for invalid documents

[1.8.2] - 2024-12-05¶

Added¶

• LLM-based entity consolidation for more accurate merging
• Priority-based conflict resolution in memory decay
• Fact order priority parsing from input text
• Concurrent evolution processing
• Full source chunk storage and retrieval

Improved¶

• Enhanced extraction prompt for more exhaustive knowledge capture
• Better handling of aliases during entity resolution
• Improved community detection summaries

[1.8.1] - 2024-12-04¶

Added¶

• LLM-based temporal extraction
• LLM-based decay reasoning
• Exhaustive community summaries
• Continuous evolution in benchmarks

Fixed¶

• Various indentation errors in evolution modules

[1.8.0] - 2024-12-04¶

Added¶

• Coreference resolution (second LLM pass)
• Alias extraction from text
• Temporal validity extraction from text
• Hybrid retrieval with alias matching
• LLM-based answer checking in benchmarks

Improved¶

• Entity resolution with alias support
• Query engine prompts for alias/temporal awareness

[1.7.0] - 2024-12-03¶

Added¶

• MemoryAgentBench evaluation support
• Synthetic benchmark for GraphMem features
• Safe concurrent query processing

Improved¶

• Evolution pipeline efficiency
• Batch ingestion performance

[1.6.0] - 2024-12-02¶

Added¶

• RAGAS-style evaluation metrics
• Skip-ingestion flag for evaluations
• Debug logging for predicted vs expected answers

Fixed¶

• MultiHopRAG dataset handling
• Evaluation accuracy calculations

[1.5.0] - 2024-12-01¶

Added¶

• Comprehensive evaluation metrics
• Token efficiency tracking
• Memory growth monitoring
• Cost per query calculation

Improved¶

• Evaluation script structure
• Profiling logs for bottleneck tracking

[1.4.0] - 2024-11-30¶

Added¶

• High-performance ingestion pipeline
• Batch embedding generation
• Async LLM extraction
• Auto-scaling worker configuration
• Benchmark utility for speedup measurement

Improved¶

• Rate limit handling with infinite retry
• Worker configuration for different hardware

[1.3.0] - 2024-11-29¶

Added¶

• Turso (libSQL) storage backend
• Native vector search for Turso
• SQLite persistence support

Improved¶

• Storage backend abstraction
• README with storage decision guide

[1.2.0] - 2024-11-28¶

Added¶

• PageRank centrality for importance scoring
• Temporal validity for relationships
• Point-in-time queries

Improved¶

• Evolution engine with PageRank integration
• Query engine with temporal awareness

[1.1.0] - 2024-11-27¶

Added¶

• Multi-tenant isolation
• Redis caching layer
• User-scoped cache invalidation

Improved¶

• Data isolation architecture
• Cache key structure

[1.0.0] - 2024-11-25¶

Initial Release¶

• Core GraphMem functionality
• Knowledge graph extraction
• Memory evolution (decay, consolidation)
• Semantic search
• Community detection
• Neo4j and in-memory storage