Memory Systems for LLM Applications
Short-term, long-term, episodic and working memory — what each one is for, what to persist, when to retrieve, and how memory fails
A working taxonomy for one of the most conflated terms in AI engineering: the conversation window (short-term), persistent facts across sessions (long-term), what-happened-last-time (episodic), and externalized scratchpads (working memory). Covers durability criteria for what to persist, eager-load-vs-retrieval-triggered design, the failure modes — staleness, injection, bloat, conflicting facts — and a concrete walkthrough of how CLAUDE.md and context compaction implement exactly this taxonomy in a real product.
Practice questions (5)
-
View →
Designing Memory for a Repeat-Visit Coding Agent
Advanced · Free -
View →
A Compacted Session Drops a Constraint the User Stated Once
Advanced -
View →
A Support Agent Re-Suggests an Already-Failed Fix
Advanced -
View →
A Memory System That Got Slow, Expensive, and Worse
Advanced -
View →
Securing an Auto-Extracting Memory Pipeline
Advanced