Knowledge Graphs in Agent Memory.md
October 15, 2025 · View on GitHub
Knowledge Graphs in Agent Memory (people may use "Memory Graph" instead of "Knowledge Graph")
Personally, there are a few keypoints that memory graph should focus on:
1. The freshness of the graph, since people's interest changes from time to time (also considered by Mem0-g and Graphiti),
2. The correctness of the entity extraction for memory, especially in multi-modal scenarios when users may use less words as instructions.
We can only guess the preference of users based on the acceptance rates of the model output such as video/figure generation.
3. The correctness of the entity matching, or the dicision of whether using the matched memory entities or not in a new scenario.
General Papers and works
- Memory [Github]
- Memary uses a graph database to store knoweldge.
- Llama Index was used to add nodes into the graph store based on documents.
- Perplexity (mistral-7b-instruct model) was used for external queries.
- ZEP: A TEMPORAL KNOWLEDGE GRAPH ARCHITECTURE FOR AGENT MEMORY (ZepAI, Arxiv 2025) [Paper] 🔥 HOT!
- Zep implements three search functions: cosine semantic similarity search (ϕcos), Okapi BM25 full-text search (ϕbm25), and breadth-first search (ϕbfs). The first two functions utilize Neo4j’s implementation of Lucene.
- Graphiti:Build Real-Time Knowledge Graphs for AI Agents [Github]
- AriGraph: Learning Knowledge Graph World Models with Episodic Memory for LLM Agents (Arxiv 2024) [Paper] [Github] hot!
- Retrieval from the AriGraph memory consists of two procedures: (1) a semantic search returns the most relevant triplets (semantic edges) and (2) an episodic search that, given extracted triplets, returns the most relevant episodic vertices
- M3-Agent: Seeing, Listening, Remembering, and Reasoning: A Multimodal Agent with Long-Term Memory (ByteDance, ACL 2025) [Paper] 🔥 HOT!
- Mem0 and Mem0-g: Building Production-Ready AI Agents with Scalable Long-Term Memory (mem0AI, Arxiv 2025) [Paper] [Github]🔥 HOT!
- Aliyun's architecture for Mem0 (discussion in Chinese) [Link]
- LightRAG: Simple and Fast Retrieval-Augmented Generation [Github] 🔥 HOT!
3.3 KG for Other Tasks in Agent
- SciToolAgent: a knowledge-graph-driven scientific agent for multitool integration (Nature computational science 2025) [Paper]