Summary
Agent Memory: Building Memory-Aware Agents is a short course created in collaboration with Oracle and taught by Richmond Alake and Nacho Martínez. It has an intermediate level, lasts 2h2m and includes 7 video lessons, 4 code examples and 1 classified task PRO.
Most agents work well at a single session but lose everything after they finish them. The course addresses long-term memory as first class infrastructure: outside the model, persistent and structured, built with Oracle AI Database, LangChain and LLM-powered pipelines.
The course teaches how to create a Memory Manager to manage different memory types and a tool semantic retrieval system that allows to scale its use without saturating the context window. The tools are stored as a procedural memory at a memory-supported warehouse and are retrieved by semantic search only when they are relevant and at the time of inference.
They also create pipelines to extract structured events from conversations, consolidate episodic memory in semantic memory and allow the agent to update and refine his memory independently. The final agent loads context prior to his initiation, uses a loop of recursive reasoning and preserves memory between sessions.
It's aimed at developers that build AI agents and they want to go beyond one-session interactions. familiarity with Python and basic concepts of LLM are recommended. The additional questionnaires and projects are included with DeepLearning.AI Pro.