Jesse Holden jesse-holden

LLM Wiki

A pattern for building personal knowledge bases using LLMs.

This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.

The core idea

Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.

Prerequisites

Use Node.js v24. Trusted Publishing does not work on earlier versions of Node.js.

Steps

Go to "Account > Access Tokens" and click "Generate Access Token".
Give the new token "read and write" persmissions to "All packages".
If you have 2FA enabled on npm (which you should), check the "Bypass 2FA" checkbox neatly hidden in the UI. Otherwise, npm will fail with an error demanding an OTP during automatic publishing.
Create the token.

Why I migrated from Bun Test to Vitest

This is not a "Bun is slow" post. In my case, Bun was often fast. I switched because I stopped trusting the failure modes in the part of my test suite that spawned real processes and local infrastructure.

Context

I was testing a Cloudflare Worker app with a mix of:

plain unit tests
tests that spawn wrangler dev

Local Code Mode for MCP

Most MCP servers just wrap CRUD JSON APIs into tools — I did it too with scim-mcp and garmin-mcp-app. It works, until you realize a tool call dumps 50KB+ into context.

MCP isn't dead — but we need to design MCP tools with the context window in mind.

This middleware does a few interesting things:

Ensures a url shape in the zustand store, where we'll store URL information.
Assumes we will be storing our url state slice in the ?state search parameter after it has been stringified and base 64 encoded.
On creation, decodes stores state from the ?state search parameter into the url slice of our store.
After each state update, updates the ?state search parameter with the new url state slice.
Sets up an event listener that listens for popstate and re-decodes the state from the URL into our store.

Next.js client-only SPA example

Made this example to show how to use Next.js router for a 100% SPA (no JS server) app.

You use Next.js router like normally, but don't define getStaticProps and such. Instead you do client-only fetching with swr, react-query, or similar methods.

You can generate HTML fallback for the page if there's something meaningful to show before you "know" the params. (Remember, HTML is static, so it can't respond to dynamic query. But it can be different per route.)

Don't like Next? Here's how to do the same in Gatsby.

Redirects in React Router v6

An important part of "routing" is handling redirects. Redirects usually happen when you want to preserve an old link and send all the traffic bound for that destination to some new URL so you don't end up with broken links.

The way we recommend handling redirects has changed in React Router v6. This document explains why.

Background

In React Router v4/5 (they have the same API, you can read about why we had to bump the major version here) we had a <Redirect> component that you could use to tell the router when to automatically redirect to another URL. You might have used it like this:

	/**
	* @description
	* Creates a frozen enumeration object where each key maps to its own string value.
	* This is useful for defining a set of constant string values that can be referenced by name.
	* This also avoids defining an object with duplicated string values manually.
	* @template {string} T
	* @param {...T} values
	* @returns {{ [K in T]: K }}
	* @example
	* const Colors = Enum('Red', 'Green', 'Blue');

	// src/routes/search/useDebouncedSearchParam.ts
	import { useEffect, useMemo, useRef, useState } from 'react'
	import { useDebouncedValue } from '@tanstack/react-pacer' // React adapter re-exports
	// Alternative import also works in examples: '@tanstack/react-pacer/debouncer'.
	// See Pacer React adapter docs. :contentReference[oaicite:5]{index=5}

	type RouteSearchFrom<RouteApi> = RouteApi extends { types: { fullSearchSchema: infer TSchema } }
	? TSchema extends Record<string, unknown>
	? TSchema
	: Record<string, unknown>