Builds a self-improving knowledge graph that enhances AI agent memory and learning capabilities.
Sponsored
MollyGraph creates a dynamic knowledge graph designed to provide AI agents with continually improving memory. Unlike traditional static vector stores, it ingests text, extracts entities and relationships through a multi-layered process, and then fine-tunes its own extraction model using LoRA. This allows the graph to autonomously learn and adapt from new data, deploying updated models only if they outperform existing ones, ensuring the graph gets smarter with every interaction. It offers parallel graph and vector retrieval for comprehensive recall and robust LLM-audited relationship validation.
Key Features
01LLM audit chain for relationship verification and reclassification
02Three-layer entity and relation extraction (GLiNER2, spaCy, GLiREL)
03Self-improving extraction via LoRA fine-tuning with A/B benchmarking
04Speaker-anchored ingestion with per-source confidence thresholds
05Parallel retrieval combining graph exact-match and vector similarity
061 GitHub stars
Use Cases
01Improving the quality of entity and relationship extraction over time without manual labeling
02Enhancing AI agent memory and context management
03Building dynamic, self-improving knowledge bases for agents
