AI Agent Memory Systems

Architect memory for AI agents — short-term context, long-term vector storage, and the retrieval strategies that decide what an agent actually remembers. Covers chunking, embedding, and recall over storage.

This skill designs agent memory architectures. It separates short-term (context window) from long-term (vector store) memory, designs chunking and embedding strategies for what to store, builds retrieval that surfaces the right memory at the right time, and adds summarization and forgetting policies so agents stay coherent across long interactions.

When to use

Use when an agent needs memory across turns or sessions — designing short-term and long-term storage and, critically, the retrieval that decides what it recalls.

Examples

Design agent memory

Short and long-term layers

Design a memory system for a support agent: short-term conversation context plus long-term vector memory of past tickets, with retrieval

Fix forgetful agents

Improve recall

My agent gives inconsistent answers across a long session — diagnose whether it's a chunking, embedding, or retrieval problem and fix the recall