📚

Chapter 2 · 10 min read

Layer 1: The Knowledge Base

Marie Kondo your agent's brain (it sparks SO much joy)

Time to build your agent's pantry. We're going to use something called the PARA method — don't worry, it's simpler than it sounds. It's basically Marie Kondo for your agent's brain. And just like KonMari, it takes an hour to set up and saves you hundreds of hours going forward.

PARA: Four Boxes for Everything

🍕 Real-life analogy

You know how Marie Kondo says to sort everything into categories? PARA is the same idea, but instead of "clothes" and "books," we have four boxes:

📦 Projects — things you're actively working on (like the box by the front door with stuff you need this week)
📦 Areas — ongoing responsibilities with no end date (like the box labeled "adulting" — bills, health, household stuff)
📦 Resources — useful reference material (like your bookshelf — you're not reading them all right now, but they're there when you need them)
📦 Archives — done or paused stuff (the storage unit — out of sight, but you can always dig it up)

The beauty of PARA is that every single piece of information in your life fits into exactly one of these four categories. There's no ambiguity. No "should this go here or there?" If it has a deadline and you're working on it → Project. If it's ongoing with no end date → Area. If it's useful reference → Resource. If it's done → Archive.

Setting It Up (Literally 2 Minutes)

terminal

mkdir -p knowledge/{projects,areas,resources,archives}

# Create a rules file so your agent knows how this works
cat > knowledge/README.md << 'EOF'
# Knowledge Base Rules

## Structure
- projects/  → Active work with deadlines (one .md per project)
- areas/     → Ongoing responsibilities (one .md per area)
- resources/ → Reference material (templates, guides, notes)
- archives/  → Completed or paused (moved here, never deleted)

## Rules
1. Every active project gets ONE file in projects/
2. When a project finishes → move to archives/ (don't delete!)
3. Ongoing responsibilities → areas/ (social media, finances, health)
4. Reference material → resources/ (templates, checklists, playbooks)
5. The nightly cron maintains everything
6. File names: lowercase-with-dashes.md (e.g., saas-mvp.md)
7. Each file starts with a # heading and current status

## For the Agent
- Read projects/ to know what we're actively working on
- Check areas/ for ongoing duties and schedules
- Search resources/ when you need templates or reference info
- Don't read archives/ unless specifically asked about old projects
EOF

That's it. Four folders and a README. Your agent now has a structured brain instead of a junk drawer.

Agent thinking...

▊

What Goes Where? (Real Examples)

Let's look at actual files you might create. Notice how each one is self-contained — your agent can open a single file and instantly have full context on that topic.

Projects: Active Work

knowledge/projects/saas-mvp.md

# SaaS MVP — TaskFlow

## Status: 🟡 Building (Started Feb 10)

## What Is This?
A task management SaaS for freelancers. Simple, opinionated, 
not trying to be Notion. $29/mo.

## Current Sprint (Feb 20-27)
- [x] Landing page live
- [x] Stripe checkout working (test mode)
- [ ] User authentication (Clerk)
- [ ] Basic dashboard UI
- [ ] Switch Stripe to live mode

## Tech Stack
- Next.js 14 (app router)
- Supabase (auth + database)
- Stripe (payments)
- Vercel (hosting)
- Tailwind CSS

## Key Decisions
- NO free tier. 14-day free trial then paid.
- Only Stripe. No PayPal (too many chargebacks).
- Mobile-responsive but no native app. Not yet.

## Links
- Repo: github.com/me/taskflow
- Staging: staging.taskflow.app
- Stripe Dashboard: dashboard.stripe.com
- Design: figma.com/file/xxx

Areas: Ongoing Responsibilities

knowledge/areas/social-media.md

# Social Media Management

## Platforms
- **Twitter/X**: @myhandle — primary platform
  - Post 3-5x/week (Mon, Wed, Fri minimum)
  - Mix: 60% value content, 30% engagement, 10% promotion
  - Best times: 9 AM and 5 PM EST
  
- **LinkedIn**: /in/myname — secondary
  - Post 1x/week (Tuesday mornings)
  - More professional tone, longer posts
  - Share project milestones and learnings

## Content Pillars
1. Building in public (what I'm working on)
2. AI agent tips and tricks
3. Indie hacker / solopreneur life
4. Occasional hot takes on tech trends

## Voice & Tone
- Conversational, not corporate
- Use analogies and humor
- Never say "leverage" or "synergy" unironically
- Short sentences. Punch. Impact.

## Metrics Tracking (weekly)
- Impressions, engagement rate, follower growth
- Top performing posts (analyze what worked)
- DM conversations started

Resources: Reference Material

knowledge/resources/email-templates.md

# Email Templates

## Cold Outreach (for partnerships)
Subject: Quick question about [their product]

Hey [Name],

I've been using [their product] for [time period] and love [specific feature]. 
I'm building [my product] which [brief description].

Would you be open to a 15-min chat about [specific collaboration idea]?

No worries if not — I'll keep being a happy customer either way.

[Name]

---

## Customer Support — Refund Request
Hey [Name],

No problem at all — refund processed. You should see it in 5-10 business days.

If there's anything specific that didn't meet your expectations, I'd genuinely 
love to know. Building this as a solo founder, so every piece of feedback 
directly shapes the product.

Either way, thanks for giving it a try.

[Name]

---

## Follow-Up (no response after 5 days)
Subject: Re: [original subject]

Hey [Name] — just floating this back to the top in case it got buried. 
Happy to adjust the ask if the timing isn't right.

[Name]

Archives: Done/Paused

knowledge/archives/old-landing-page-v1.md

# Landing Page V1 (ARCHIVED)

## Archived: Feb 20, 2026
## Reason: Replaced with V2 (interactive design)

## What It Was
Static single-page landing. Hero + features + CTA.
Conversion rate: 2.3% (below 3% target).

## What Worked
- Headline "Your AI agent has amnesia. Let's fix that." — good hook
- Social proof section pulled 40% of scroll engagement

## What Didn't
- Too much text above the fold
- CTA was "Learn More" (weak — changed to "Start Building")
- No pricing preview = too many tire-kickers

## Lessons for V2
- Show price early (filters out non-buyers)
- Use interactive elements (not just text walls)
- Lead with pain, not features

🎯 The Payoff

When your agent needs to work on your SaaS project, it doesn't dig through a giant memory file. It opens knowledge/projects/saas-mvp.md and instantly has full context — tech stack, current sprint, key decisions, all relevant links. Like opening the right cookbook to the right page instead of flipping through every book you own.

The "How Do I Know If It's Working?" Test

After setting up your knowledge base, try this: start a brand new conversation with your agent and ask it about your project. Without re-explaining anything. If the agent can tell you:

✓ What you're working on
✓ What the current status is
✓ What the next steps are
✓ What tech stack you're using

...then congratulations, your knowledge base is working. If it can't, the file probably needs more detail or your agent's configuration isn't pointing to the right directory.

Common Mistakes

🚫 Files that are too long

If a project file is over 200 lines, it's too long. Split it. Your SaaS project doesn't need the API documentation in the same file as the sprint plan. Keep each file to one screen's worth of reading.

🚫 Files with no status

Every project file needs a status indicator at the top. 🟢 Live. 🟡 Building. 🔴 Blocked. Your agent should be able to glance at the first line and know if this project is active or stalled.

🚫 Never archiving anything

Projects finish. When they do, MOVE the file to archives/. Don't just leave it in projects/ with a "DONE" note. Your agent reads the projects/ folder to know what's active. Old files = confusion.

🚫 Organizing resources by date instead of topic

Don't name files "notes-feb-22.md." Name them by what they contain: "email-templates.md," "competitor-analysis.md," "pricing-research.md." Your agent searches by topic, not date.

Pro Tips

💡 Pro Tip: Let Your Agent Maintain Its Own Knowledge Base

Add this to your AGENTS.md: "After completing any significant task, update the relevant knowledge base file with learnings, decisions, and status changes." Your agent will start maintaining its own brain. You just review the changes occasionally.

💡 Pro Tip: Use the README as a Router

Your agent reads knowledge/README.md first to understand the structure. Put "routing hints" in there: "For anything about payments → projects/saas-mvp.md. For content ideas → areas/social-media.md." This helps the agent find the right file faster.

Advanced: Cross-Referencing Between Files

Here's something most people miss: your knowledge base files can reference each other. In your project file, mention which area it relates to. In your area file, list active projects. This creates a web of context that helps your agent navigate intelligently.

knowledge/projects/saas-mvp.md (with cross-refs)

# SaaS MVP — TaskFlow
## Related: areas/social-media.md (for launch promotion)
## Related: resources/pricing-research.md (pricing decisions)
## Related: resources/competitor-analysis.md (feature priorities)

Think of it like hyperlinks in your brain. When your agent reads your project file and sees those references, it knows exactly where to look for supporting context. No guessing, no searching.

Platform-Specific Setup

🤖 OpenClaw / Claude

Point your AGENTS.md to read from the knowledge/ directory. Claude loves markdown — keep files clean and well-structured with headers. Use the README.md as a routing file.

🦜 LangChain / LlamaIndex

Use a vector store to index your knowledge/ folder. Chunk by file (not by paragraph) — each file is already a coherent unit. Re-index on file change via a watcher.

🔧 Custom Agents

Build a simple tool that lists files in each PARA folder and reads them on demand. Don't dump everything into the system prompt — let the agent pull what it needs.