Guides

Agent Memory Architecture Part 2: Episodic vs Semantic vs Procedural Memory

The three agent memory types — episodic, semantic, and procedural — explained through the real production code of ClaudeClaw, a 24/7 multi-agent system, so your agents stop starting from scratch every session.

A practitioner's guide to solving agent 'amnesia' by giving agents the three distinct memory types production systems need: episodic (what happened, and when), semantic (facts), and procedural (how to behave). It walks through real code from ClaudeClaw — a multi-agent system running 24/7 on the Claude Agent SDK on Apple Silicon — with concrete schemas, a one-page decision framework, and a local-LLM (SQLite + Ollama) implementation path. Real code and tradeoffs, not hypotheticals.

$ tar -tzf agent-memory-architecture-part-2-episodic-vs-semantic-vs-procedural-memory.tar.gz
  • **Practitioner guide** (~4,000 words, 10 sections) — plain-English definitions followed immediately by production code walkthroughs: the data/procedural/scout.md injection pattern, the Memory v2 canonicalizer contract, and the WATCHMAN sweep ProbeFinding schema with its timestamp-as-episodic-anchor design
  • Memory Type Decision Framework — one-page reference card: use case → memory type → ClaudeClaw implementation → SQLite-based local-LLM alternative
  • Local-LLM section — concrete SQLite schemas for Ollama users who can't use hosted memory services
  • Honest failure modes — what breaks when you pick the wrong memory type, sourced from real operational post-mortems
$27

one-time purchase

Instant download after purchase
📧Download link sent to your email
🔄7-day download access
14-day money-back guarantee
View refund policy

## goes-well-with

From the same shelf

## not-ready-to-buy

Take the field notes instead

One practical write-up a week from the same workbench these kits come from — plus reader pricing when new kits ship.