TutorialsThree Questions + Two Tools
What is a second brain / AI knowledge base? It's not a cloud drive or a chatbot — the core logic is Karpathy's LLM Wiki 'compiler' paradigm. Meet WorkBuddy (does the work) and Obsidian (shows the structure).
What This Note Is
This note is not a lecture — it's a hands-on manual you follow at home.
Tonight, 30 minutes: run the minimal loop of "build → ingest → query." After that, 10 minutes a day keeps the library growing. No coding — everything is "chat + prompts."
Core principle: AI does the first 4 steps, you do the last 1 — you state the goal and make judgments; AI does the heavy lifting.
You need:
- A computer (Windows or macOS)
- Two tools installed: WorkBuddy (workhorse), Obsidian (viewer, free)
- 10–30 materials you use most (industry articles, work docs, study notes; 3 is enough for tonight)
What Is a "Second Brain / AI Knowledge Base"
One-sentence definition:
AI Knowledge Base = your materials + AI-organized knowledge cards + a maintenance ruleset — a personal knowledge system that keeps growing.
In plain terms: give AI a "dedicated reference room" — it only holds your materials, only answers from your materials. Things you've saved and learned finally have a place to "be stored and retrieved."
Three Layers
| Layer | Name | Owner | One-liner |
|---|---|---|---|
| ① raw/ | Source layer | You | Curated sources (articles/docs/PDFs), read-only = ground truth |
| ② wiki/ | Knowledge layer | AI | Generated cards/summaries/index, AI maintains = value-added |
| ③ Schema | Rules layer | Ruleset | A rules file telling AI how to maintain = AI's manual |
Each layer has its job:
- You manage sources: pick materials and put them in raw/
- AI manages organizing: digests into cards, cross-links them
- Rules manage maintenance: constrain AI — facts need sources, originals can't be changed, one thing at a time
What It Is NOT
Not a Cloud Drive
| Cloud Drive / Folder | AI Knowledge Base |
|---|---|
| Store but don't edit — dead materials | AI digests, distills, cross-links |
| Manual sorting, gets messier | AI auto-organizes, gets richer |
| Find by remembering filenames | Ask in plain language, AI searches |
| No relation to chat AI | AI only answers from your materials, with sources |
Not a Regular Chat AI
| Chat AI (Doubao/DeepSeek/Wenxin) | AI Knowledge Base |
|---|---|
| Answers from its own knowledge | Only searches your materials |
| Doesn't know your materials | Remembers everything you've ingested |
| Gone after the chat | Keeps growing |
| No source attribution | Every claim traceable to source |
In short: Doubao is a "talking assistant"; a second brain is a "working, accumulating digital employee."
Why It's Worth It
Chat AI's weakness isn't capability — it's no persistence, no accumulation. After the chat, it's gone; it can't recall what you saved six months ago.
Traditional knowledge management (Zettelkasten, GTD, mind-map archives…) relies on manual maintenance: the more materials, the harder organizing and searching get, so people give up.
AI Knowledge Base hands "maintenance" to AI: you curate materials, ask good questions, and make judgments; AI handles organizing, cross-referencing, and upkeep.
The Core Metaphor
Remember this:
- Traditional RAG (AI searches and answers on the fly) = interpreter: runs from scratch every time, accumulates nothing.
- LLM Wiki (knowledge "compiled" into the library) = compiler: compiled once, continuously updated, grows over time.
We're learning the compiler path.
Source: Andrej Karpathy (OpenAI founding member, former Tesla AI Director, Stanford PhD) proposed the LLM Wiki knowledge base paradigm.
Meet Your Two Tools
WorkBuddy: The AI Digital Employee
It doesn't just chat — it can work hands-on in your authorized folder: read materials, build the library, digest into cards, retrieve on demand, produce outputs.
Three modes (how much AI does):
| Mode | One-liner | When to use |
|---|---|---|
| ASK | Answer only, don't touch files | Thinking through ideas, worried about changes. Safest |
| PLAN | Give you a plan first, act after you approve | Many steps, feeling unsure |
| CRAFT | Do it directly when you say what you want | Task is clear. Most actions in this course use it |
Five elements of a task (all prompts derive from this formula):
Task = Goal (what problem to solve) + Input (which materials) + Output (what format) + Constraints (what not to do) + Acceptance (what counts as done)
Core entry points (beginners start here):
- New Task: top-left, start a new piece of work
- Task input box: middle-bottom of the new task page, type your request
- Working directory: below or bottom-right of input, limits read/write to a folder (set your knowledge base folder here)
- Add context: + button next to input, upload files, paste screenshots, reference docs
- Execution: conversation area / task details, see what it's reading and doing
- Result area: usually on the right, preview delivered docs/tables
Obsidian: The Readable Knowledge Interface
What it is: a knowledge management tool using plain text (Markdown) notes, with bidirectional links connecting notes into a personal knowledge network. Core features are free.
Why you need it: WorkBuddy shows you conversations and file changes; Obsidian shows you structure at a glance — file tree, clickable links, graph view — how knowledge connects is crystal clear.
How they share files: storage and viewing are separate — WorkBuddy and Obsidian open the same folder. Files are plain .md text; even if you stop using Obsidian, your files are always your own.
Tool Comparison (Why This Combo)
| WorkBuddy | Obsidian | |
|---|---|---|
| Role | AI digital employee: hands-on + accumulating | Knowledge dashboard: view + trace |
| Strength | Out-of-the-box, built-in AI, read/write/produce, task acceptance criteria | Free, local files, bidirectional links, graph view, files are yours |
| Handles | Read, build, digest, query, retrieve, write | Display structure, click links, view graph, trace sources |
| Use when | All daily operations (send prompts) | When you want to see structure or verify sources |
In short: Obsidian views, WorkBuddy works.
Why not use Obsidian for everything?
- Obsidian needs you to connect AI models (install plugins) for digest/split/index capabilities — better for advanced users
- WorkBuddy has built-in AI, ready to use, no setup
- Obsidian core is free; WorkBuddy is pay-per-use
- Conclusion: WorkBuddy as the AI brain (processing), Obsidian as the visual interface (display) — not that Obsidian can't, just that setting up AI on it is more hassle
Common Worries
| Question | Answer |
|---|---|
| Do I need to code? | No. Everything is chat + prompts |
| Is it safe? | Files are local; WorkBuddy only works in your authorized directory |
| Is it expensive? | Obsidian is free; WorkBuddy is pay-per-use, minimal cost at start |
| Is it complicated? | Follow Steps 1–5 tonight, about 30 minutes total |
Build an AI Knowledge Base
Build your second brain from zero in 30 minutes — WorkBuddy + Obsidian dual-tool × 7-prompt toolkit × 6 industry scenarios.
7-Prompt Toolkit (Copy & Go)
Build, ingest, query, retrieve, restock, write, maintain — 7 prompts, each with 'when to use → prompt → how to adapt → how to verify.' Copy, paste, run.