Skip to main content
← Back to sources

Andrej Karpathy Proposes 'LLM Knowledge Base' Architecture Using Evolving Markdown Libraries as RAG Alternative

Published 2026-04-08AI Engineering PracticesHigh

Summary

Andrej Karpathy has shared a new architecture concept called 'LLM Knowledge Base' that proposes replacing traditional Retrieval-Augmented Generation (RAG) pipelines with a curated, evolving library of markdown documents maintained by AI. The approach envisions LLMs continuously organizing, updating, and distilling knowledge into structured markdown files that can be fed directly into model context windows, eliminating the need for vector databases, embedding models, and retrieval infrastructure

Alignment: New signal not yet covered
Related Positions: ai-infrastructure-strategy.md, agentic-workflows.md, ai-assisted-development-tooling.md, enterprise-ai-delivery.md
Related Partnerships: glean.md
karpathyrag-alternativeknowledge-managementmarkdown-knowledge-basecontext-windowsagentic-maintenanceenterprise-knowledgellm-architectureai-engineering-practicesinfrastructure-simplification
Andrej Karpathy Proposes 'LLM Knowledge Base' Architecture Using Evolving Markdown Libraries as RAG Alternative — Intelligence — Agentic Developer Tools Radar · Signal