I Built a Durable AI Knowledge Base with Markdown and Git

Short summary

The article describes a three-layer architecture for AI knowledge bases using Markdown and Git: raw sources for provenance, maintained knowledge pages for evolving synthesis, and agent schemas for rules. It argues that search alone doesn't create memory—different document types need different change contracts. The approach complements RAG by persisting conclusions and contradictions rather than rediscovering them each query.

• •Three-layer model: raw sources, maintained knowledge, agent schema—each with distinct change contracts
• •RAG retrieves fragments but doesn't persist synthesis; a maintained wiki stores conclusions, relationships, and unresolved disagreements
• •Directory structure (10-inbox through 90-archive) defines how each document type ages and who can rewrite it

Generated with AI, which can make mistakes.