WorkBuddy Knowledge Base: 4-Step Loop to Turn Scattered Docs into Your AI Tutor

Clearing Three Misconceptions

The author begins by addressing three common misconceptions about knowledge bases. First, a knowledge base is not a cloud drive; cloud drives only pile up files into a graveyard, whereas a knowledge base lets you and AI share a single living artifact so that today's saved version becomes tomorrow's starting point. Second, it is not a traditional note-taking app; notes are written for yourself, while a knowledge base is a workspace where humans and agents collaborate on plans, drafts, and shared tables under one roof. Third, do not over-engineer the directory structure upfront; the author has seen many people (including their past self) design library-like hierarchies with three-level categories and five-level tags, only to abandon them. The ultimate goal is not collection but enabling AI to pick up your accumulated context and continue the work.

Four-Step Closed Loop: Save, Reference, Edit, Return

Opening WorkBuddy's sidebar and clicking "Knowledge Base" reveals a four-step cycle that can be completed in one evening:

Save. Preserve a valuable draft from a chat session into "My Documents" with a descriptive filename such as 20260907-knowledge-base-tutorial-draft. Avoid generic names like "New Document 3" that become meaningless later.

Reference. Back in the knowledge base, select the document and click "Add to Task." This tells the AI: do not guess from scratch; continue from this artifact.

Edit. Issue a single, focused instruction—for example, "Condense section two to under 400 words and add three data points." A critical detail: AI modifications on the document remain in a pending state until you explicitly accept them; generation does not equal finalization, and confirmation rights stay with you.

Return. Save the refined result back into the knowledge base. Next time you start, pull from the library instead of excavating chat history.

Completing this loop once transforms scattered, ephemeral chat outputs into a growing, reusable knowledge asset.

Three Containers, Each with a Distinct Role

The knowledge base supports three artifact types, and the author provides a mnemonic derived from personal trial and error: Edit content in Markdown, edit data in CSV, edit presentation in HTML.

Markdown (MD) for narratives: proposals, weekly reports, article bodies—it manages the story flow.

CSV for structured data: topic lists, schedules, data logs—it serves as the single source of truth; changing a number means editing only here.

HTML for renderable pages and charts—it handles presentation and can directly visualize CSV tables as charts.

These containers can cross-reference: documents embed tables, pages read data. The knowledge base thus evolves from a pile of isolated files into a small, interconnected system.

Three Pitfalls the Author Already Stepped In

Uploaded files are often read-only. For tables and documents you intend to edit daily, create new MD or CSV files directly inside the library rather than preparing them locally and uploading. Local-to-upload suits archiving, not active work.

Task-generated artifacts default to "Local Artifacts." If you do not manually save them into "My Documents," switching devices or waiting a few days makes them disappear again. Save into the library, not the conversation.

Sensitive data hygiene. Phone numbers, raw client spreadsheets, unmasked records—scrub them before ingestion. AI permissions will not overreach, but if you feed dirty data, it will read it faithfully.

On the credit system: storage, organizing directories, retrieval, and manual editing consume no credits; only AI-powered processing, analysis, and generation do. The author finds this design reasonable—browsing your own library is free, while AI labor is metered.

Ready-to-Copy Directory Structure

Create the following folders under "My Documents": 00-Materials — dump collected fragments. 03-Drafts — works in progress. 04-Published — final versions that have shipped. 99-Templates-and-Standards — prompt templates, writing guidelines, and other reusable assets.

The structure is simple but effective. The author's 100,000-word handbook was organized exactly this way. Once set up, you can instruct the AI: "Using the standards in 99, wrap up the draft in 03," and it actually finds the right context. That moment reveals the true value of a personal AI tutor: not how smart it is, but that it remembers what you have practiced.

Closing Thought

Having migrated from Obsidian and felt the frustration of "beautiful tools, empty systems," the author is convinced by WorkBuddy's knowledge base because it does not demand you become a knowledge-management master before you deserve a knowledge base. Run the save–reference–edit–return loop for one night; let time handle the rest.