🚀 Getting Started
From zero: installation, configuration, SOUL.md template, first task. Must-read for beginners. Read →
OpenClaw Field Notes
Why this exists
Section titled “Why this exists”I’ve been running OpenClaw for three months. Scaled from one agent to five, running across two Macs, executing tasks 24/7.
The number of problems I hit along the way exceeds the amount of code I’ve written.
Online content about OpenClaw falls into two categories: official docs (what you can do) and Twitter screenshots (what others did). Almost nobody writes about: what actually goes wrong in practice, and how to fix it.
This handbook fills that gap.
Reading path
Section titled “Reading path”Different starting points for different stages:
🟢 Day 0 — Just installed OpenClaw, not sure what’s next
Section titled “🟢 Day 0 — Just installed OpenClaw, not sure what’s next”→ Start with Getting Started — step-by-step from zero to your first working agent.
🟡 Day 7 — Single agent running, want to add more
Section titled “🟡 Day 7 — Single agent running, want to add more”→ Read Architecture — when to split? How to go from 1 to 5? Hardware choices?
🔴 Day 30 — Multi-agent running for a while, problems appearing
Section titled “🔴 Day 30 — Multi-agent running for a while, problems appearing”→ Read Debug Playbook and Workflow Recipes — specific problems, specific solutions, specific configs.
🚨 Something broke — Agent down / API errors / nothing happening
Section titled “🚨 Something broke — Agent down / API errors / nothing happening”→ Jump to Emergency Triage — three steps to locate the problem.
Navigation
Section titled “Navigation”Architecture
The complete evolution from 1 agent to 5. Hardware, role design, communication, monitoring. Read →
Debug Playbook
Model selection disasters, API quota burns, agents going silent for 6 hours… each with full context and fix. Read →
Workflow Recipes
QA gates, task queues, heartbeat monitoring, memory optimization, cross-gateway coordination. Read →
Opinions
Contrarian views on AI agents: why organization > product, why agents don’t need memory. Read →
About
What this handbook is, who’s writing it, community resources. Read →
Setup at a glance
Section titled “Setup at a glance”- Hardware: MacBook Air (coordinator) + Mac Mini (workers)
- Agents: 5 — COO (dispatch), Coder (engineering), Research (research), Marketing (marketing), QA (QA)
- Communication: Discord (inter-agent) + Telegram (human interface)
- Models: Kimi 2.5 primary, OpenAI Codex fallback, MiniMax budget tier
- Monthly cost: ~$30-50 (down from $200+)
- Tested on: OpenClaw 2026.3.x (specifically 2026.3.2)
Everything here is based on real operational data. Not demos. Not theory.
Community resources
Section titled “Community resources”Can’t find an answer? Try these:
- OpenClaw official docs: docs.openclaw.ai — authoritative feature documentation
- OpenClaw GitHub Issues: github.com/anthropics/claude-code/issues — search known issues and solutions
- GitHub Discussions: github.com/anthropics/claude-code/discussions — community discussions, best practices
- Discord community: OpenClaw official Discord — real-time help
How this handbook differs from other resources
Section titled “How this handbook differs from other resources”| Official Docs | Twitter/Posts | This Handbook | |
|---|---|---|---|
| What you can do | ✅ | ❌ | ❌ |
| What others did | ❌ | ✅ | ❌ |
| What goes wrong | ❌ | ❌ | ✅ |
| How to fix it | Partial | ❌ | ✅ |
| How much it costs | ❌ | ❌ | ✅ |