Your Pi-hole Isn’t Catching Everything, Plus everything I missed in Part 1

A comprehensive follow-up to Pi Hole and Unbound setup, addressing common questions, pitfalls, blocking strategies, and remote access options.

---

1. Summary

This video demonstrates how to build a persistent, production-grade AI co-working system using OpenClaw as the agent runtime and Obsidian as the structured control interface for system files, memory, and workflow orchestration.

The core problem addressed: Most users install OpenClaw but fail to create a durable, structured context system—leading to fragmented prompts, inconsistent agent behavior, and cognitive overload.

The solution: A two-machine architecture where:

• Mac Mini runs the agent (24/7 execution environment)
• MacBook Pro is the human control interface
• A centralized user.md acts as the single source of truth
• System files are modularized but injected cohesively
• Obsidian Sync keeps both machines synchronized
• Plugins allow rendering and managing non-markdown files

Primary technologies:

• OpenClaw
• Obsidian
• Obsidian Sync
• Data Files Editor plugin
• HTML Reader plugin

---

2. Detailed Step-by-Step Breakdown

Phase 1 — Install Core Components

Step 1: Install Obsidian

• Download from: https://obsidian.md/
• Install on:
  • Mac Mini
  • MacBook Pro

Step 2: Install OpenClaw

(Transcript does not specify installation method.) Assumption: Standard Setup

• Clone repository:

git clone <openclaw-repo-url>
cd openclaw

• Install dependencies:

npm install

or

pip install -r requirements.txt

(depending on implementation)

• Run agent:

openclaw start

or

npm run dev

---

Phase 2 — Create the Core System File Architecture

Inside your OpenClaw workspace, structure folders as follows:

/workspace
│
├── system/
│   ├── user.md
│   ├── identity.md
│   ├── tools.mmd
│   ├── agents.mmd
│   ├── soul.md
│
├── memory/
│   └── memory.md
│
├── tasks/
├── outputs/
└── dashboards/

---

Phase 3 — Build the Most Important File: user.md

This is your single source of truth.

Contents of user.md should include:

• Business model
• Revenue streams
• Target audience
• Voice and tone
• Content strategy
• Current projects
• Constraints
• Definitions of success
• Active experiments

Example structure:

# Business Overview

## Brand Positioning
...

## Audience
...

## Monetization
...

## Current Priorities
...

## Writing Voice
...

Why this replaces context profiles

Old approach:

• Multiple fragmented context profiles
• Hard switching between them
• Inconsistent injection

New approach:

• One canonical user.md
• Always loaded at runtime
• Agent inherits persistent worldview

---

Phase 4 — Modular System Files

Each file has a defined concern:

identity.md

Defines agent personality and operating style. Example:

You are Proxy.
You are concise.
You prioritize leverage.

tools.md

Defines:

• Tool permissions
• CLI commands
• External integrations
• API capabilities

agents.md

Defines:

• Sub-agents
• Delegation hierarchy
• Responsibilities

memory.md

Stores:

• Persistent learnings
• Corrections
• User preferences
• Ongoing task state

---

Phase 5 — Two-Machine Architecture

Machine 1: Mac Mini

• Runs OpenClaw continuously
• Handles agent execution
• Acts as headless AI server

Machine 2: MacBook Pro

• Human interaction layer
• Edits markdown files
• Delegates tasks
• Reviews outputs

Both machines access the same Obsidian vault.

---

Phase 6 — Configure Obsidian Sync

Inside Obsidian:

1. Settings → Sync

2. Enable Obsidian Sync

3. Select Vault

4. Log in on both machines

5. Enable:

   • Sync Markdown
   • Sync attachments
   • Sync hidden files (important for system files)

This ensures:

• Edits to user.md propagate instantly
• Agent reads updated context
• Memory stays aligned

---

Phase 7 — Install Required Plugins

Data Files Editor Plugin

Allows editing of non-standard file formats.

Install:

• Settings → Community Plugins
• Browse → Search: “Data Files Editor”
• Install → Enable

HTML Reader Plugin

Allows rendering HTML dashboards in Obsidian.

Search:

• “HTML Reader” or “obsidian-html-plugin”
• Install and enable

---

Phase 8 — Make Non-Markdown Files Visible

Obsidian normally hides:

• .json
• .mmd
• .html

Use plugins to:

• View tool configuration files
• Render HTML dashboards
• Edit structured data files

---

Phase 9 — Render HTML Dashboards

Create:

/dashboards/status.html

Example:

<h1>Agent Status</h1>
<p>Current Task: Newsletter Draft</p>

Open inside Obsidian using HTML Reader plugin.

This turns Obsidian into:

• Monitoring panel
• Agent cockpit
• Delegation dashboard

---

Phase 10 — Delegation Workflow (Live Demo Concept)

Example task delegation to agent “Proxy”:

1. Open /tasks/new-task.md
2. Write:

# Task
Draft outline for next newsletter on AI co-working systems.

Constraints:
- 1200 words
- Strategic tone
- Include 3 examples

1. Agent loads:

   • user.md
   • identity.md
   • tools.mmd
   • memory.md

2. Output saved to:

/outputs/newsletter-draft.md

---

3. Key Technical Details

Persistent Context Injection

Agent runtime loads:

BASE_CONTEXT =
    identity.md
  + user.md
  + memory.md
  + tools.mmd

Order matters:

• Identity defines behavior
• User defines worldview
• Memory defines continuity
• Tools define capability

---

Folder Separation Principles

---

System Requirements (Assumption: Typical Setup)

• macOS 13+
• 16GB RAM recommended
• Node.js 18+ (if JS-based)
• Python 3.10+ (if Python-based)
• Stable internet (for Sync)

---

4. Pro Tips

1. Treat user.md as a Living Document

Update weekly. Refine based on:

• Misaligned outputs
• Strategy shifts
• Product evolution

---

2. Avoid Context Fragmentation

Do NOT:

• Create multiple competing user files
• Hard-code business logic into prompts

Keep one canonical source.

---

3. Use Memory Intentionally

Only store:

• Persistent truths
• Corrections
• Behavioral adjustments

Do NOT store:

• Temporary task details

---

4. Use HTML Dashboards for:

• KPI tracking
• Task queues
• Agent status visualization

---

5. Separate Execution from Editing

Mac Mini = compute MacBook Pro = cognition

This reduces:

• Distraction
• Performance bottlenecks
• Workspace clutter

---

5. Potential Limitations / Warnings

⚠️ Sync Conflicts

If both machines edit same file:

• Obsidian creates conflict copies
• Agent may load wrong version

Mitigation:

• Assign write ownership
• Avoid simultaneous editing

---

⚠️ Plugin Dependency Risk

If plugin breaks:

• HTML dashboards may fail
• Data file visibility may disappear

Keep backups of:

• System files
• Vault

---

⚠️ Context Bloat

Overloading user.md:

• Slower reasoning
• Diluted signal
• Increased hallucination risk

Keep it strategic and structured.

---

⚠️ Hidden File Sync Issues

Ensure:

• “Sync hidden files” enabled
• .mmd files are not excluded

---

6. Recommended Follow-Up Resources

• Official Obsidian documentation
• OpenClaw GitHub repository
• Obsidian plugin development docs
• Knowledge management methodologies (PARA, Zettelkasten)
• Prompt engineering and context architecture research

---

7. Suggested Books (5)

1. Building a Second Brain — Tiago Forte

Explains structured knowledge management systems. Directly supports designing an Obsidian vault that scales with AI collaboration.

2. Designing Data-Intensive Applications — Martin Kleppmann

Deep dive into state management, synchronization, and data architecture—critical for understanding multi-machine setups and consistency.

3. The Pragmatic Programmer — Andrew Hunt & David Thomas

Covers system thinking, automation workflows, modularity, and maintainability—aligned with building structured AI systems.

4. Clean Architecture — Robert C. Martin

Teaches separation of concerns and system layering—mirrors identity/tools/memory modular structure used in this workflow.

5. Deep Work — Cal Newport

Not technical in code, but directly relevant to the execution philosophy behind separating machines and designing frictionless co-working systems.