Prompt Kit: AI Knowledge Architecture — Wiki vs. Database vs. Hybrid

This kit helps you make the most consequential design decision in your AI workflow: how your AI remembers what it learns. Based on the write-time compilation (Karpathy wiki) vs. query-time retrieval (Open Brain) comparison, these prompts help you choose your architecture, design the schema that governs it, run the actual synthesis process, audit for failure modes, and blueprint a hybrid system.

How to use this kit

Start with Prompt 1 if you haven't decided on an approach yet — it produces a personalized recommendation based on your actual situation. Prompt 2 is essential for anyone building a wiki system — the article makes clear that the schema is the highest-leverage document in the entire setup, and most people will underinvest in it. Prompt 3 is the operational engine — use it every time you feed new sources into a wiki. Prompt 4 is for anyone with an existing knowledge setup that might be silently drifting or hiding contradictions. Prompt 5 designs a complete hybrid architecture if you want both approaches working together. Run all prompts in a thinking-capable model like ChatGPT, Claude, or Gemini for best results. Prompts 3 and 4 work best when you can share actual files or content with the AI.

Prompt 1: Knowledge Architecture Advisor

Job: Diagnoses your specific knowledge management situation and recommends whether to build a wiki, a structured database, or a hybrid — with concrete implementation steps.

When to use: Before you start building anything. Also useful if you have an existing system that feels wrong but you can't articulate why.

What you'll get: A personalized architecture recommendation with rationale, a risk profile for your choice, specific failure modes to watch for, and a prioritized implementation checklist.

What the AI will ask you: What you use AI for, whether you're solo or on a team, the volume and types of information you manage, how you typically query your knowledge, what tools you already use, and what's frustrating about your current setup.

Prompt 2: Wiki Schema & Editorial Policy Designer

Job: Designs the schema file — the editorial policy that tells the AI how to organize, synthesize, and maintain your wiki. The article identifies this as the single highest-leverage document in the entire system.

When to use: Before you start building a wiki, or when your existing wiki feels disorganized and you realize the AI needs better instructions for how to structure knowledge.

What you'll get: A complete, ready-to-use schema document that defines page types, cross-referencing rules, contradiction handling, what to include vs. exclude, and the AI's editorial standards.

What the AI will ask you: Your domain/topic area, what types of sources you'll be feeding in, what connections matter most to you, and how you plan to use the wiki.

Prompt 3: Wiki Synthesis Agent

Job: Acts as the AI maintainer for a Karpathy-style wiki. You give it a source, it reads the source, and produces or updates wiki pages — integrating new knowledge against what already exists, flagging contradictions, building cross-references.

When to use: Every time you add a new source to your knowledge base. Paste this prompt at the start of a session, then feed it your source material and any existing wiki pages that might need updating.

What you'll get: New or updated wiki pages with synthesis, cross-references, contradiction flags, and an updated index — all in markdown ready for Obsidian or any note app.

What the AI will ask you: Your wiki schema (or it will help you define key parameters), the new source material, and any existing wiki pages relevant to the new source.

Prompt 4: Knowledge Base Drift & Contradiction Auditor

Job: Audits your existing knowledge base — whether it's a wiki, a collection of notes, or a database of entries — for the specific failure modes the article identifies: wiki drift (confident prose that's quietly wrong), hidden contradictions, stale syntheses, and gaps where important connections are missing.

When to use: Periodically (monthly is a good cadence), or whenever you suspect your knowledge base might be telling you a cleaner story than reality supports. Especially important for teams where multiple people or agents are contributing.

What you'll get: A diagnostic report identifying specific instances of drift, contradictions, staleness, and gaps — with severity ratings and recommended fixes.

What the AI will ask you: A sample of your knowledge base content (wiki pages, notes, database entries) and context about how recently different parts were updated.

Prompt 5: Hybrid Knowledge System Blueprint

Job: Designs a complete hybrid architecture — structured database as source of truth, compiled wiki as the readable layer — tailored to your specific situation, tools, and workflow.

When to use: When you've decided you need both structured storage and compiled synthesis, and you want a concrete blueprint for how the two layers connect, what tools to use, and how the compilation process works.

What you'll get: A full system design document with architecture diagram (text-based), folder/database structure, compilation workflow, tool recommendations based on what you already use, and an implementation sequence you can start building today.

What the AI will ask you: Your current tools and setup, what types of knowledge you manage, your technical comfort level, whether you're solo or team, and your priorities.