// Track

Agent Knowledge Systems

Build the Memory Layer Where Corrections Compound

Build persistent memory for AI agents — the capture, index, recall, and lifecycle architecture that makes agent quality compound across sessions instead of resetting. Memory taxonomy and provenance, dedupe-gated write paths, progressive-disclosure recall, namespaced authority, hygiene (supersede, expire, verify-before-trust), budgeted session hydration, and the health metrics that prove the loop is working.

Recommended: Complete RAG in Production first

10 lessons~82 min total

Lessons are shown in recommended order. Complete them in sequence for the best experience — or jump to any lesson.

Lesson 586·7 min read

Why Agents Forget — and What a Knowledge System Is

Sessions are amnesiac by design — a knowledge system is the architecture that makes corrections compound instead of evaporate.

Lesson 587·7 min read

The Memory Taxonomy: Episodic, Semantic, Procedural, Working

Memory type is not decoration — it determines lifecycle policy, confidence floor, and when a memory should die.

Lesson 588·7 min read

The Write Path: Search Before You Store

The write path is a dedupe gate, not a log — search before store, supersede instead of append, soft-delete instead of overwrite.

Lesson 589·8 min read

Indexing & Recall: Progressive Disclosure

Search returns pointers, not memories — fetch full content only for the 1-2 that matter, because every recalled token competes with task context.

Lesson 590·8 min read

Namespaces & Authority: Who May Know What

A memory store without access control is a single shared truth for every agent — and one wrong write poisons the fleet.

Lesson 591·9 min read

Memory Hygiene: Supersede, Expire, Verify

Memories record what was true when written. The store does not receive world-change events. Verify before you act.

Lesson 592·8 min read

Session Context Assembly: The Hydration Budget

Every recalled token competes with the task itself. Load the index cheaply; fetch depth only for what the session needs.

Lesson 593·9 min read

The Compound Learning Loop: Corrections Become Rules

Every correction that evaporates at session end is a correction you will make again — the compounding loop converts one-time fixes into permanent behavior change, with receipts.

Lesson 594·9 min read

Memory Ops: Health, Drift & Observability

A process can run while the store rots — health means the artifact is working, not just the service; four metrics tell you the difference.

Lesson 595·10 min read

Capstone: Build a Knowledge OS

Assemble the canonical store, dedupe-gated write path, progressive recall, namespaced authority, hygiene jobs, budgeted hydrator, and health metrics into one Knowledge OS — then trace how every lesson becomes a load-bearing component.