---
title: "Google OKF + Claude : Why We Stopped Using RAG"
source_url: https://youtu.be/l46NJXUL4PM
captured_at: 2026-07-06
author: "Cloud Codes"
contributor: "ชิดชนก"
platform: youtube
---
# Google OKF + Claude : Why We Stopped Using RAG

youtube video by **Cloud Codes**, 9m45s, uploaded 2026-07-04. URL: https://youtu.be/l46NJXUL4PM
Tags: claude code, google okf, open knowledge format, ai agents, rag, vector database, agentic search, llm memory, markdown, system design, software engineering, cloud codes, embeddings, knowledge graph, artificial intelligence, machine learning, tech tutorial, anthropic, prompt engineering, silicon valley, llm wiki, wikipedia llm, llm agent, claude code free, claude code free unlimited, claude code vs 

## Shared in LINE Note+
- 2026-07-06 18:11 by ชิดชนก

## Description
Why do AI agents suffer from amnesia, and why did Anthropic completely rip out the local Vector Database from Claude Code? Discover how to give your AI agents perfect, version-controlled long-term memory using Google's new Open Knowledge Format (OKF) and Agentic Search instead of traditional RAG pipelines.

In this video, Cloud Codes breaks down the ultimate AI memory stack. We explain why relying on embeddings, cosine similarity, and Vector Databases causes confident hallucinations, and why "Agentic Search" (letting an LLM literally grep and read local files) is infinitely more accurate. 

We dive deep into Google OKF (v0.1)—a standard that replaces complex databases with a simple folder of Markdown files and YAML frontmatter. We show you exactly how to wire an OKF bundle directly into Claude Code (using CLAUDE.md and MEMORY.md) to create a portable, human-readable knowledge graph that your agent can read, update, and submit via GitHub Pull Requests. Finally, we give an honest verdict on when you still actually need a vector database.

⏱️ TIMESTAMPS:
00:00 - The Problem: AI amnesia
00:16 - The Flaw: Why vector databases fall short
00:48 - The Pivot: Exact search beats fuzzy guessi