Skip to content
AI|Course

Managing data · Advanced

Engineer Persistent Memory for AI Agents

Agents are stateless by default. Learn how to give them short-term and long-term memory that stays accurate, affordable and safe: what to store, how to retrieve it, and how to prove it works.

7 lessons · 13 hr 45 min · Free

You will learn to

  • ▸Explain the working, episodic, semantic and procedural memory types and when each applies
  • ▸Keep long threads inside the context window with trimming, compaction and notes
  • ▸Design a long-term store with extraction, consolidation, expiry and deletion
  • ▸Compare tiered, scored and temporal-graph memory architectures
  • ▸Evaluate memory quality and defend it against leaks and poisoning

Before you start

  • ·Comfortable with Python and SQL
  • ·You have built a tool-calling agent or chatbot
  • ·Helpful: the GraphRAG course, for the graph-based lessons

Syllabus

  1. 01Why agents forgetStateless models, finite context, and what bad memory costs.45 min
  2. 02A taxonomy of agent memoryWorking, episodic, semantic and procedural memory.1 hr
  3. 03Short-term memory and context managementCheckpoints, trimming, compaction and structured notes.2 hr
  4. 04Long-term memory storesWrite, read and maintenance paths, with a Postgres schema.2 hr 30 min
  5. 05Memory architecturesMemGPT and Letta, Generative Agents, and Zep with Graphiti.2 hr
  6. 06Evaluation, privacy and safetyBenchmarks, test design, leaks and memory poisoning.1 hr 30 min
  7. 07Capstone projectShip an assistant that measurably improves by remembering.4 hr