← Back to CoursesAgentic AI: Advanced

Neuroanatomy Explorer

Drag to rotate · scroll to zoom · click regions to explore

View
Loading 3D model…

Click a region
to explore it

Memory Deck

Flip each card and rate whether you knew it. Your score is saved.

Term
Definition

Deck complete — score saved.

Match the Pairs

Match each term to its definition. Finish the board to earn your score.

All matched — score saved.

The Agent Loop in 3D

Watch a thought travel through Perceive → Plan → Act → Observe. Drag to rotate, scroll to zoom, click a node.

Click a node to read its definition.

Memory, State, and Retrieval-Augmented Agents

Manual: General · Subject: Agentic AI

Examine how agents store, update, and retrieve information for persistent competence.

Memory as a Computational Resource

Forms of Memory

Agentic systems commonly use short-term working memory, episodic memory, semantic memory, and procedural memory. Working memory tracks the current task; episodic memory stores past interactions; semantic memory holds facts; procedural memory encodes reusable skills.

Retrieval-Augmented Acting

When context windows are limited, agents rely on retrieval to fetch relevant items from a store using similarity, metadata, recency, or task-specific filters. This enables scalable long-term operation without repeatedly re-sending all prior information.

⚠️

Memory Is Not Truth

Retrieved content can be stale, incomplete, or incorrect; agents should treat memory as evidence, not ground truth.

Memory Design Questions

What to store
Only durable, useful, or safety-relevant information
When to retrieve
At task boundaries or uncertainty spikes
How to index
By embeddings, keywords, structure, or timestamps
How to update
With consolidation, decay, or human review

Which memory type is most directly used to store facts about the world?

Why is retrieval useful for agentic systems with long tasks?

Memory Failure Modes

Common errors include over-retrieval, under-retrieval, false consolidation, leakage across tasks, and confirmation bias. Research-grade systems often attach provenance, confidence scores, and deletion policies to stored memories.

What is a strong reason to attach provenance to stored agent memories?