Designing Semantic Memory for Multi-Agent Systems with Python
summary
Argues that semantic memory is a data engineering problem and shows design choices that actually change results: layered memories—short-term conversational state, episodic event logs, declarative and procedural memory—with lifecycle ops (extract, summarize, supersede, TTL) and hybrid retrieval combining structured filters plus semantic signals. Implements a Python multi‑agent travel planner using LangGraph, LangChain integrations, Azure OpenAI, and Azure Cosmos DB, demonstrates session vs per‑turn persistence, checkpointed workflows for traceability, and an agent memory toolkit that separates reasoning from storage.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 3:16Memory taxonomy & architecture
- 12:14Neuroscience of agent memory
- 15:23Structured memory documents
- 16:27Salience scoring for recall
- 18:55Memory lifecycle & compression
- 20:42Retrieval & session persistence
- 22:49Shared memory coordination
- 28:45Control flow in agents
- 30:29Conflict resolution in memory
- 36:04Conversation as code
- 37:29Memory challenges in coding
- 41:41Memory toolkit demo
- 43:50Import/export conversations
- 44:33Summary
related talks
18mArchitecting Agent Memory: Principles, Patterns, and Best Practices
29mThe AI Agent Memory Landscape
51mMemory Masterclass: Make Your AI Agents Remember What They Do!
15mBuilding Evolving AI Agents Via Dynamic Memory Representations Using Temporal Knowledge Graphs
25mWhat does it take to build long-term state for AI agents?
28mAgentic Memory
32mOpenSearch Agentic Memory: Intelligent Context for AI Agents
01h 49mBuilding Memory Augmented Agents with MemoRizz