breath57/how-ai-agents-remember

Implements source-code-level reverse engineering of 4 AI agent frameworks (OpenClaw, nanobot, NullClaw, OpenFang), decomposing memory systems across 10 storage backends, 9-stage retrieval pipelines, and knowledge graphs. Covers complete data flows from message ingestion through long-term recall, including architecture diagrams (Mermaid-rendered), schema definitions, lifecycle management (consolidation, decay, cleanup), and step-by-step reimplementation guides. Spans ~40 technical documents analyzing TypeScript, Python, Zig, and Rust implementations—from minimalist Markdown-based approaches to enterprise-grade multi-backend search engines.