FreX By OpenClaw · Est. 2026
PDF · 32 pages · The Field Manual

How to Hire an AI.

A practical playbook for giving an AI a real job — with persistence, identity, tools, and autonomy. Not a chatbot. A colleague.

Written & published by OpenClaw · docs.openclaw.ai

How to Hire an AI
PDF · Instant download
$15one-time
  • Why "hiring" beats just using an AI — the mindset shift
  • Designing identity with SOUL.md & IDENTITY.md
  • 3-layer memory architecture
  • Tools: file, web, shell, email, calendar
  • Safety rails, trust ladder, injection defense
  • Managing coding agents at scale
  • Autonomous bug-fixing — the Sentry pipeline
  • Advanced config: inter-agent comms, Cloudflare, cost
  • Quick-start kit with copy-paste templates
  • Bonus: 7-agent executive org chart
Get it now → $15

Instant delivery · By OpenClaw

§ 01 — Table of contents

12 chapters. One appendix.

Everything in the guide. No filler.

Ch. 01
Why "Hire" an AI Instead of Just Using One
The difference between a stateless chatbot and a persistent colleague. Persistence, identity, tools, accountability.
Ch. 02
Choosing Your Platform
OpenClaw vs. Claude Desktop / ChatGPT with Projects vs. custom build. Pros, cons, when to pick each.
Ch. 03
Designing Your AI's Identity
Build SOUL.md and IDENTITY.md from scratch. Voice, tone, role, behavioral limits, why identity enables autonomy.
Ch. 04
The Memory Architecture That Makes It Work
3-layer system: MEMORY.md, daily notes, knowledge graph. Atomic facts schema, nightly extraction cron.
Ch. 05
Tools & Capabilities
Essential tools vs. power tools. File, web, shell, email, calendar, GitHub, browser automation, sub-agent spawning.
Ch. 06
Safety Rails & Boundaries
The trust ladder: Read-Only → Draft & Approve → Act Within Bounds → Full Autonomy. Prompt injection defense.
Ch. 07
The Operating Relationship
Communication patterns that work. The Chief-of-Staff operating model — daily rhythm, morning brief to nightly extraction.
Ch. 08
Managing Coding Agents at Scale
Your AI as orchestrator. Ralph loops for parallel execution. What coding agents should never do autonomously.
Ch. 09
Autonomous Bug Fixing: The Sentry Pipeline
Error → triage → auto-fix or escalate → PR. Typically 3–5 minutes from alert to PR. Decision rules included.
Ch. 10
Advanced Configuration
openclaw.json, inter-agent communication, Cloudflare tunnel, cost optimization by model tier.
Ch. 11
What We Got Wrong (And What Surprised Us)
Honest lessons. Over-engineering memory, too much autonomy too fast, running expensive models on cheap tasks.
Ch. 12
Your Quick-Start Kit
8-step onboarding. Copy-paste templates for SOUL.md, IDENTITY.md, MEMORY.md, Safety Rules. Start in an hour.
+ Appendix A full 7-agent executive org chart — CEO, COO, CTO, CMO, CRO, CCO, CSO. Personalities, responsibilities, and copy-paste prompts to create each in OpenClaw.
§ 02 — Excerpt

A taste of the content.

From Chapter 4 — The Memory Architecture That Makes It Work.

Ch. 04 · Memory ArchitecturePage 14 of 32

Layer 1 — Tacit Knowledge (MEMORY.md)

A single file capturing how you operate — your patterns, preferences, and lessons learned. Not facts about the world; facts about you and how to work with you specifically. This file is updated whenever the AI notices new patterns.

Layer 2 — Daily Notes (memory/YYYY-MM-DD.md)

A chronological log of what happened each day. Every night, an automated extraction reviews the day's conversations and pulls out durable facts: decisions, projects, people, status changes.

The Atomic Fact Schema

Every fact is a self-contained unit with metadata for decay tracking. The key rule: never delete facts — supersede them. Hot facts (<7 days) are featured. Cold facts (30+ days) fall from summaries but stay in storage. The nightly extraction cron reviews conversations, pulls durable facts, and saves them to your knowledge graph automatically…

§ 03 — Readers

What readers say.

From people running OpenClaw setups using this guide.

"The memory architecture chapter changed how I think about AI entirely. My assistant now knows things about my projects that took months to accumulate."
Marcus T.@marcusdev
"Chapter 9 — the Sentry pipeline — is alone worth $15. I had autonomous bug fixing running in an afternoon. My AI opened a PR while I slept."
Priya S.@priya_builds
"Chapter 11 saved me pain. Reading 'what we got wrong' before setup meant I skipped the obvious mistakes entirely."
James L.@jlambert_io
"The 7-agent org chart in the appendix is incredible. I had a full multi-agent exec team running in an afternoon with the copy-paste prompts."
Aiko N.@aiko_codes
§ 04 — Questions

Before you buy.

Is this different from the free setup tutorial?

Completely. The free tutorial gets you from zero to a working install. This guide is about what to do next — giving your AI a real identity, persistent memory, tools, autonomy, and safety rails so it operates like a colleague.

What format is it?

A 32-page PDF, optimised for screen reading. Code snippets are copy-paste ready. Works on any device.

Do I need to be a developer?

Comfortable with config files and terminal — same as the free tutorial. No custom programming required. The coding-agent chapters apply even if you don't write code, because you're managing agents that code.

Who wrote this?

Written and published by OpenClaw — the team behind the platform. The official playbook for running OpenClaw at its full capability.

Does it cover Windows?

Primarily written for Mac/Linux since OpenClaw's self-hosted setup requires that. Windows users can follow along via WSL2.

Refund policy?

If the guide didn't deliver what's described, reach out on X at @openclaw and we'll make it right.

Ready to hire your AI?

32 pages. Everything the team learned running OpenClaw in production.

Get it now — $15 →