Adding RAG: Store Your Own Judgment and Grow Your AI

A classic bottleneck when using AI coding agents like Claude Code is loading a 32KB index of custom notes at the start of every session, only to watch the AI repeat the exact mistakes those notes were written to prevent. Whether it is reminding the tool to capture screenshots in the English UI or keeping articles concise, each correction gets logged as a memory note. Yet, the missing link is timing: when a situation arises mid-task, the agent rarely goes back to search through past records on its own.

This limitation is not unique to Claude Code; any AI coding agent utilizing saved notes must actively decide to search them, and that decision step is frequently skipped. To bridge this gap, retrieval-augmented generation (RAG) can be integrated so that a search runs automatically on every message sent, fetching the most relevant past notes and handing them over before the AI answers.

Stock photo for illustration only, not from the actual event

This memory recall mechanism operates via a Claude Code hook—a script executing automatically at fixed moments, currently upon every message sent. Nothing undergoes retraining; rather, what changes is the context handed to the AI. Every recorded decision and correction becomes reference material available the next time a similar situation surfaces, compounding your personal judgment into the AI's workflow.

Long notes are split into 700-character segments, scoring each piece individually so deeper details remain discoverable. To prevent noise from overwhelming unrelated chats, the system evaluates the score gap between top hits rather than relying on a rigid threshold, ensuring irrelevant queries trigger no extra context injection.