Agentic AI
Intermediate
Memory Architectures for Long-Context Agents
Implement episodic, semantic, and working memory buffers for persistent AI agents.
Course Overview
Explore stateful memory architectures that allow AI agents to maintain coherent long-term user context without overflowing token budgets or leaking cross-session state.
What You Will Learn
Implement multi-tiered agent memory: working, short-term, and long-term
Store and retrieve episodic memories using vector search and timestamp decay
Summarize long conversations without losing critical user preferences
Prevent cross-tenant memory leakage in multi-user agent systems
Measure context retention accuracy across multi-session interactions
Tools & Technologies Used
Python 3.11
Qdrant
Redis
LangGraph
FastAPI
Structured Curriculum
2 Modules · 4 Lessons · 180 Minutes Total
Module 1
Module 1: Memory Abstractions & Storage Layers
2 lessons
Module 1: Memory Abstractions & Storage Layers
Structure working memory buffers and persistent vector storage.
-
Episodic vs. Semantic Memory Models
Map cognitive memory concepts into software data structures.
-
Context Compression & Rolling Summarization
Build adaptive summarization hooks that preserve key facts while truncating raw history.
Module 2
Module 2: Retrieval & Multi-Session Continuity
2 lessons
Module 2: Retrieval & Multi-Session Continuity
Query historical memories intelligently during live agent execution.
-
Time-Decayed Vector Memory Retrieval
Combine recency scoring with semantic similarity for optimal memory recall.
-
Multi-Tenant Memory Isolation & Privacy
Implement tenant-isolated namespace keys to prevent cross-user data leakage.
Practical Project & Capstone Outcome
🚀 Capstone Project
Stateful Long-Term Executive Assistant Engine
Build an AI executive assistant that maintains long-term memory of user preferences, meeting notes, and project history across multi-week sessions.
Prerequisites
- Python intermediate proficiency
- Basic understanding of vector stores
Intended Audience
- AI Developers building conversational agents with long-term persistence
- Product Engineers designing personalized AI companions and assistants
Instructor Information
M
Marcus Vance
Course Author & Industry Expert
Marcus Vance is a Principal AI Systems Engineer with extensive experience in memory compression algorithms and vector storage.
Frequently Asked Questions
Which memory stores are integrated during the course?
We use Redis for high-speed working memory and Qdrant for long-term episodic vector storage.