Day One: Your 45-Minute Setup
Less time than an episode of The Office. More life-changing.
Enough theory. Let's build this thing. Here's your exact 45-minute setup — step by step, nothing skipped. I've done this dozens of times and refined it to the minimum viable setup that actually works.
Four folders and a README. Your agent's organized brain.
Whatever you're working on right now. Give your agent context.
Start with 5 communication preferences. You'll add more over time.
Copy the config from Chapter 5. Your agent's night shift starts tonight.
List 2-3 things for your agent to check periodically.
Give your agent personality and boundaries. This is its identity.
Tell your agent about the knowledge base and security model.
Step-by-Step Walkthrough
Step 1: Create the PARA Structure (5 min)
# Create the folder structure
mkdir -p knowledge/{projects,areas,resources,archives}
mkdir -p memory
# Create the knowledge base README
cat > knowledge/README.md << 'EOF'
# Knowledge Base (PARA)
- projects/ → Active work (one .md per project)
- areas/ → Ongoing responsibilities
- resources/ → Reference material
- archives/ → Completed work
Rules:
1. One file per project/area
2. Finished projects → archives/
3. File names: lowercase-with-dashes.md
4. Each file starts with status emoji: 🟢 live, 🟡 building, 🔴 blocked
EOF
echo "✅ PARA structure created"Step 2: Write Your First Project File (10 min)
Open your editor and create a file for whatever you're working on right now. Use this template:
# [Project Name] ## Status: 🟡 Building (Started [Date]) ## What Is This? [2-3 sentences explaining the project] ## Current Priority - [ ] [Most important task right now] - [ ] [Second most important] - [ ] [Third] ## Tech Stack / Tools - [List what you're using] ## Key Decisions Made - [Any important decisions so far] ## Links - [Repo, staging URL, dashboards, docs]
Don't overthink this. Spend 10 minutes, not 2 hours. You can always add more later.
Step 3: Create tacit.md (10 min)
# What I Know About My Human ## Communication Style - [How do you like messages formatted?] - [Short or detailed?] - [Formal or casual?] ## Pet Peeves - [What annoys you in AI interactions?] - [What should your agent NEVER do?] ## Work Patterns - Timezone: [Your timezone] - Most productive: [When?] - Don't bother me: [When?] ## Decision Style - [Do you want options or recommendations?] - [How much autonomy should the agent have?] ## Technical Preferences - [Languages, frameworks, tools you prefer]
Step 4: Set Up Nightly Consolidation (5 min)
openclaw cron add \
--name "Nightly Memory Consolidation" \
--cron "0 2 * * *" \
--tz "YOUR_TIMEZONE" \
--session isolated \
--message "Review today's daily note. Extract key \
decisions and learnings. Update knowledge base files. \
Update tacit.md with any new preferences discovered. \
Log what you consolidated." \
--model "sonnet"Step 5: Create HEARTBEAT.md (5 min)
# Heartbeat Checklist - [ ] Any urgent emails? - [ ] Calendar events in next 2 hours? - [ ] Any social mentions/DMs? - If nothing: reply HEARTBEAT_OK - Late night (11 PM - 8 AM): only urgent stuff
Step 6: Write SOUL.md (5 min)
# Who You Are You are an operator, not a help desk. Be genuinely helpful, not performatively helpful. Skip "Great question!" — just answer the question. Have opinions. Disagree when you think I'm wrong. Be resourceful before asking. Try to figure it out. Ship fast, iterate later. You're a guest in someone's life. Be respectful. Be the assistant you'd actually want to talk to.
Step 7: Update AGENTS.md (5 min)
Add the memory system rules, security model, and daily notes protocol from the previous chapters. (The Pro tier gives you copy-paste-ready configs for this.)
The Verification Test
After completing all 7 steps, run this test:
Your agent should greet you with context — mentioning your project, current status, etc. If it says "Hi! How can I help?" with no context, something's not connected.
Without re-explaining anything, ask "What's the status of my project?" Your agent should be able to answer from the knowledge base file.
Ask for a summary of something. Does the output match your communication preferences from tacit.md? If you said "no tables" and you get a table, the tacit file isn't being read.
Let the nightly consolidation run once. The next morning, check if the knowledge base was updated and a daily note exists.
What to Do If Something Isn't Working
Check your AGENTS.md — does it tell the agent where to look? Add: "On startup, read knowledge/tacit.md and memory/YYYY-MM-DD.md for context."
Run openclaw cron list to verify it's registered. Check the timezone is correct. Check logs with openclaw cron logs.
The tacit.md file might not be in the agent's read path. Explicitly tell it: "Read knowledge/tacit.md before responding." If that works, add it to AGENTS.md permanently.
That's the Blueprint tier complete. You now have everything you need to run a functional AI agent with persistent memory. The next tier — Pro — gives you battle-tested configs you can copy-paste, plus real case studies of agents making money.
Share this chapter
Chapter navigation
8 of 36