Tutorials
Hands-on guides for running OpenClaw AI Agent. Start from scratch and level up step by step.
Not sure where to start?
Take the 2-minute AI capability check โ find your gap and what to learn next.
About OpenClaw
Why You Need OpenClaw: When ChatGPT and Gemini Aren't Enough
You're already using 5 AI tools, but your information is still scattered everywhere. See how OpenClaw helps you connect them all.
Token Economics: Understanding How AI Bills You (So Your Invoice Doesn't Explode)
What are tokens? How do different models charge? Learn to precisely control your AI usage costs and get the most value for your money.
The Full Panorama of AI Tech Evolution: From Transformer to Swarm Intelligence, All in One Read
A 2017 paper changed the world. In 10 minutes, understand the full arc of AI/LLM technological development, and learn why every technology behind OpenClaw exists.
AI Tool Landscape: The Five Walls Framework for Understanding AI Evolution
Why do more AI tools mean more confusion? Because you're missing a map, not more tutorials. Use the Five Walls framework to understand every leap in AI capability.
Is OpenClaw Right for You? Walk Through These Five Layers First
A lot of people ask: should I use OpenClaw? There's no universal answer โ but there is a map. This five-layer framework helps you see where you are now, what your next step looks like, and what each layer actually costs.
Environment Setup
CLI Beginner's Guide: What Is the Command Line and Why Can't AI Live Without It?
The command-line interface (CLI) is the hands and feet of OpenClaw. Understand how CLI works, common tools, and why it's the most powerful way to operate in the AI era.
Sign Up for GitHub: One-Click with Google, or Fill the Form (with Device Verification)
Create a GitHub account step by step: the fastest way is one-click sign-up with your Google account, or you can fill in the email / password form manually. Both paths walk you through device email verification and onto your Dashboard. Many tutorials start by assuming a GitHub account โ this one gets you past that first.
Google API Key Complete Guide: Essential for Connecting OpenClaw to Google Drive / Gmail
A step-by-step guide to applying for a Google Cloud API Key โ solving the most common sticking point when connecting OpenClaw to Google services.
Sign Up for Kaggle: Google Login + Phone Verification, Done in One Go
Create an account on Kaggle, the world's largest data-science community: click Register on the homepage, sign in fast with your Google account, and finish with SMS phone verification to unlock full features. Many AI/data-analysis tutorials start by assuming a Kaggle account โ this one gets you past that first.
OpenRouter Sign-Up: One Key for Every Big Model (Free Models Included)
OpenRouter is an LLM gateway โ one API to reach every major model. Sign up in minutes with a Google account, get an API key, and use a pile of free models. This walks you from homepage sign-up, to choosing an account type, to grabbing your key, to finding free models โ ready to plug into Make, n8n, OpenClaw, or your own code.
A Tour of GitHub Developer Settings: Where to Manage Fine-grained Tokens and SSH/GPG Keys
Not sure where GitHub keeps your Personal Access Tokens, SSH/GPG keys, and GitHub Apps? This guide walks the pages under Settings โ Developer settings, focuses on the safer Fine-grained tokens, and explains the passkey re-verification for sensitive actions.
Choose Your AI Brain: A Complete Comparison of 4 LLM Options
OpenClaw needs an LLM to work. Understand the differences between ChatGPT subscriptions, OpenRouter, and various APIs to pick the best option for you.
Get an Ollama Cloud API Key: Open Models as a Backup Provider
Ollama isn't only for running models locally โ it also offers cloud models and API keys. This shows you how to sign up for Ollama and create a cloud API key under Settings, so you can use Ollama as another LLM provider in Make, n8n, or code โ a fallback for when your primary provider is down or rate-limited.
AI Model API Key Guide: OpenAI, Anthropic, Google, OpenRouter โ All in One Place
Step-by-step guide to applying for AI model API Keys. No matter which provider you choose, follow along and get the key that powers OpenClaw.
Installation & Deployment
Why You Shouldn't Install OpenClaw on Native Windows
OpenClaw's core runs on CLI tools and Markdown files. Learn why macOS and Linux are the best habitats for your lobster, and what Windows users should do instead.
Hermes Agent Quick Start | ollama launch hermes
Launch Hermes AI Agent with one command on macOS: automatic install, cloud model selection, and setup โ experience an agent with cross-session memory and 70+ built-in skills in minutes.
Installation Guide: Choose Your Platform
OpenClaw installation entry page. Choose the installation tutorial for your operating system.
Complete WSL Guide: Give Your Windows Linux Superpowers
WSL (Windows Subsystem for Linux) is the best way for Windows users to install OpenClaw. From installation to configuration to daily use โ all in one guide.
Hermes Agent Quick Start | Windows + WSL2
Windows guide: run Hermes AI Agent on WSL2 + Ollama with one command. Experience cross-session memory and cloud model speed โ no API key required.
Install OpenClaw on macOS: A Complete Step-by-Step Guide
A step-by-step guide to installing OpenClaw on Mac, covering Homebrew, Python environment, dependencies, and your first launch.
Install OpenClaw on Windows: A Complete Step-by-Step Guide
A step-by-step guide to installing OpenClaw on Windows 10/11, covering Python, WSL, dependencies, and your first launch.
Deploy OpenClaw to the Cloud: One-Click Zeabur Deployment + KimiClaw Quick Start
Don't want to deal with local setup? Deploy OpenClaw to Zeabur with one click, or try KimiClaw for an instant cloud AI Agent experience.
Ollama + OpenClaw Quick Start | macOS
A macOS-only guide: use Ollama's free cloud quota to get OpenClaw talking in minutes โ no API key setup required at all.
Ollama + OpenClaw Quick Start | Windows
A Windows-only guide: using WSL + Node 22 and Ollama's free cloud quota to get OpenClaw talking โ no API key setup required at all.
Deploy a Static Site / PWA to GitHub Pages: From Creating the Repo to Going Live
Deploy your front-end-only site (HTML/JS, PWA) to GitHub Pages for free: create a repo, push your code, enable Pages in Settings, and let GitHub Actions build and publish it automatically. Full screenshots, plus build-failure and domain troubleshooting.
Docker + n8n on Windows: A Complete Pitfall Guide
A beginner-friendly guide to installing Docker, WSL2, and n8n on Windows. Covers the 5 most common failure points with step-by-step fixes, including corporate IT restrictions.
Docker + n8n on Mac: Complete Setup Guide
The simplest way to run n8n with Docker on macOS. No WSL2, no slash confusion, no line-ending issues โ just three commands.
Essential Dev CLI Tools on Mac: Homebrew, GitHub CLI, Firebase CLI, Antigravity CLI
Install the developer command-line tools you actually use on Mac in one go. Lay the foundation with Homebrew, then add GitHub CLI, Firebase CLI, and Google's Antigravity CLI โ with login and troubleshooting.
Basic Usage
First Launch of OpenClaw: Set Up Your API Key and Hear AI's First Words
Installation done โ now let's bring OpenClaw to life. Follow the setup wizard to initialize everything and get AI to respond within 3 minutes.
Your First Skill: Build Your First Automation Task in 5 Minutes
Your OpenClaw can already chat, but its real power lies in Skills. Follow along and turn a manual task into a one-click automation.
How to Set a System Prompt in ChatGPT, Claude, Gemini & Grok
Stop re-explaining who you are every time you open an AI. A system prompt (custom instructions) is set once and applies to every chat after. This guide walks you through setting one in ChatGPT, Claude, Gemini, and Grok, with a copy-paste starter template.
Model Configuration & Switching: Let OpenClaw Auto-Select the Best AI Model
Different tasks call for different models. Learn to configure Providers, model routing, and Fallback mechanisms โ save money and boost efficiency.
What Is caffeinate? Keep Your Mac Awake So AI Agents Don't Drop All Night (and How to Stop It)
caffeinate is a built-in macOS command that prevents your Mac from sleeping, so long-running AI Agents like OpenClaw and Hermes don't drop when the computer sleeps. This article explains in plain words what caffeinate is, what each of the five -dimsu flags does, common patterns, how to stop it with killall caffeinate, and the traps people fall into.
Core Features
OpenClaw Skill Complete Guide: Teach AI Repeatable Workflows
Skills are OpenClaw's core feature. Master Skills, and your AI can complete in one click what used to take 30 minutes.
OpenClaw Agent Complete Guide: Build Your AI Double
The Agent is the soul role of OpenClaw โ it understands your intent, auto-selects Skills, and even decides what to do next on its own.
OpenClaw Soul Complete Guide: Give Your Agent Memory, Personality, and Growth
Soul is OpenClaw's most powerful and unique feature. It turns your Agent from just a tool into an AI partner that truly 'knows you.'
MCP Protocol: AI's USB Port โ Plug-and-Play Tools
Model Context Protocol lets AI connect to any tool. Learn how MCP works and why it's the backbone of OpenClaw.
RAG Explained: The Memory Technique That Stops AI from Making Things Up
AI hallucinates and its knowledge goes stale. RAG technology makes AI look up information before answering โ turning 'guessing' into 'citing sources.' Learn how OpenClaw's Memory system uses RAG.
Integration & Automation
Your First Make Automation: Auto-Save New Gmail into Google Sheets
Use Make (formerly Integromat) to build your first no-code automation: whenever Gmail gets a new email, automatically write the sender, subject, and body as a row in a Google Sheet. This covers picking a ready-made template, authorizing the Google connection, and configuring the trigger and write modules โ the foundation for adding AI auto-classification later.
Telegram Integration Complete Guide: Turn OpenClaw Into Your Personal AI Assistant
From creating a bot with BotFather to connecting OpenClaw โ turn Telegram into your anywhere, anytime AI entry point.
Google Cloud: Create a Project, Enable APIs, Configure the OAuth Consent Screen
When tools like Make/n8n need sensitive Google permissions (read/write Gmail, Sheets, Drive), 'Sign in with Google' isn't enough โ you must create your own Google Cloud project, enable the matching APIs, and set up the OAuth consent screen and client credentials. This breaks down that slightly daunting path step by step.
Create a LINE Login Channel and Add a LIFF App: Turn Your Web Page Into a Mini App Inside LINE
Create a LINE Login channel in the LINE Developers console, add a LIFF App under the LIFF tab, and grab your LIFF ID and LIFF URL. A 6-step walkthrough with screenshots โ plus what LIFF is for, where it fits, and the spots people get stuck.
Auto-Classify Gmail with an LLM in Make: Just Add One AI Module
Level up the previous 'Gmail โ Google Sheets' automation: insert an OpenRouter LLM module in the middle so AI reads each email, auto-tags it 'support / sales / spam / urgent,' and writes the label into the sheet. This shows you how to wire the OpenRouter connection, pick a free model, and write a solid classification prompt.
What Is n8n? The Enterprise Automation Pain Points It Solves
Employees copy-pasting between systems all day, pulling reports by hand, SaaS tools billing per run and blowing the budget โ n8n wires all that busywork together with visual workflows. Duck Editor explains why companies are obsessed with it.
Create Your First Firebase Project
Want to add cloud login or data sync to a front-end-only site? Step one is creating a project in the Firebase Console. This article takes you from the welcome page to the project overview, and explains what Firebase does for you when you have no backend.
Build an AI Cloud Ordering System with Gemini: From Menu to Kitchen Display in 4 Steps
Write zero lines of code. Four prompts let Gemini build your Google Apps Script backend, an automated form, and a live kitchen ticket display. The point isn't just the code โ it's learning to handle real dev problems: mismatched columns, missing permissions, and forgotten redeploys.
Enable Google Sign-In in Firebase
Add 'Sign in with Google' to your site using Firebase Authentication: open Authentication, pick the Google provider, and set a project support email. Screenshots included, plus the most common support-email warning and how to clear it.
Register a Web App in Firebase to Get the SDK Config
Register your site as a Web app under your Firebase project and grab the firebaseConfig SDK snippet so your front-end can connect. Screenshots included, plus whether the apiKey is actually a secret and where to find the config again later.
Fix "Missing or insufficient permissions": Deploy Firestore Security Rules with the Firebase CLI
Wired a front-end-only site to Firestore but the cloud connection test throws "Missing or insufficient permissions"? That's Firestore's default lock-down rules. This article deploys your firestore.rules with the Firebase CLI, verifies in the Console, and checks authorized domains โ with full screenshots.
Connect Grok to GitHub: One Sentence, and the AI Creates the Repo and Pushes the Files (Tested on the Free Tier)
Connect GitHub under 'Skills & Connectors' on grok.com, then type one plain sentence โ 'write a website introducing a CLI tool and upload it to a github repo' โ and Grok creates the repo, writes the files, and pushes them. Includes how to scope the GitHub permissions safely, plus free-tier test results from 2026-07-21.
Connect ChatGPT to GitHub: How to Set It Up, and What the Free Tier Actually Did
Find GitHub in ChatGPT's Plugins page, install it, and finish the authorization โ three minutes end to end. But what can it do once connected? Tested 2026-07-21 on the free tier: connecting and reading works; getting it to actually create a repo and push files did not. Here are the steps and the honest results.
Automation Workflows
The Reconciliation Workflow: Two Ledgers, Matched in One Click
Bank statement against internal books, platform payouts against orders, receivables against payments โ normally you compare them line by line, and you still miss things. This automation workflow lets AI do the matching and flags whatever doesn't line up. Open the interactive demo and see how it works.
The Reporting Workflow: Turn Numbers into a Dashboard in One Click
Every week and every month you pull the data by hand, paste it into Excel, work out the period-over-period change, build a chart, then squint at it looking for what went wrong. This automation workflow computes the metrics, finds the trend, flags the outliers and draws the dashboard for you. Open the interactive demo and see how it works.
The Inbox Structuring Workflow: Scattered Messages into a Tracking Table
Reading emails one by one and copying the to-dos into a list by hand? This automation workflow pulls out the counterparty, the action, the due date and the amount from your emails and messages, then builds a structured tracking table sorted by urgency. Open the interactive demo and see how it works.
The Document Diff Workflow: Two Versions, and Exactly What Changed
A contract comes back marked up, a quote gets revised, an SOP is updated โ reading clause by clause to find what moved, and what now works against you, is exhausting and easy to miss. This automation workflow compares the two versions clause by clause, marks additions, deletions and edits, and points out the risky ones. Open the interactive demo and see how it works.
Knowledge & Advanced
Analyze a CSV in Looker Studio: From Connector to Data Source
Google's free Looker Studio (formerly Data Studio) can turn a CSV file straight into an interactive dashboard. This walks you through the crucial first step: pick the 'Upload CSV' connector, authorize Google Cloud Storage, create a dataset, upload the file, and confirm the field schema โ laying the groundwork before you build any charts.
Prompt Engineering: Making AI Understand You and Give the Answers You Want
Master the core techniques of prompt engineering โ from role-setting to chain-of-thought โ and improve the same model's output quality by 10x.
Which AI Tool Should I Use? 5 Questions to End the Guessing
Wrong tool, wasted effort. Use this decision framework to find the right AI tool for any task in 5 questions โ no more guessing or random Googling.
Build a Personal Knowledge Management (PKM) System with OpenClaw
Say goodbye to information overload. Use OpenClaw to automatically collect, organize, and review knowledge โ make your second brain actually work.
AI Reasoning Techniques Explained: Chain-of-Thought, ReAct, and Tree of Thoughts
How did AI learn to 'think'? From Chain-of-Thought to Tree of Thoughts, discover the techniques that boost AI reasoning by 5x, and how OpenClaw uses them.
Multi-Agent Collaboration and Swarm Intelligence: When AI Learns Teamwork
A single Agent has limited abilities, but a group of Agents working together can solve complex problems. Learn about Multi-Agent collaboration patterns, swarm intelligence concepts, and how OpenClaw implements multi-role coordination.
The Missing Infrastructure of AI Memory: Why Stronger Models Make Agents More Likely to 'Run in Circles'
Memory isn't an add-on feature for AI โ it's the core infrastructure that transforms an Agent from a 'tool' into a 'partner.' This article won't walk you through a paper; instead, it helps you think clearly: if you're designing an AI Agent, what problem is the memory layer actually solving?
Duck Editor's Notes
AI Now Operates the Web for You: From Comet and Atlas to a Full Computer-Use Browser Guide
Browsers no longer just 'display web pages' โ they click buttons, fill forms, and finish tasks themselves. How do Perplexity Comet and ChatGPT Atlas differ? This guide helps you pick the right one.
Three AIs, One Shippable LINE Booking System: The Duck's Collaboration Log
Instead of asking 'which AI is strongest,' divide the labor: use Claude to write the dev plan, use Antigravity CLI (a Gemini agent) to actually build the code, then use Claude Opus to hunt bugs and iterate. This is the Duck's real log of using this 'AI division of labor' to build a working LINE course-booking PWA โ it's about the mindset, not a step-by-step tutorial.
AI Using Tools On Its Own Is Becoming the BaselineโSo You Don't Need to FOMO
Stuck on the 'Tools' dimension in the quiz? Don't panic. On July 9 OpenAI shipped GPT-5.6, merged ChatGPT and Codex into one desktop app, and launched ChatGPT Work, a cross-app agentโwhich proves one thing: letting AI call tools and access your data on its own is shifting from 'you wire it up yourself' to a built-in baseline. Here's what that actually means for a non-technical you, and where to spend your effort instead.
OpenClaw or Hermes for Beginners? First Figure Out Whether You Want to Learn or to Get Work Done
People keep asking me which Agent tool a beginner should pick. My answer: don't rush to pick a tool. First get clear on whether you want to learn the fundamentals, or to boost your productivity. Those two things call for effort in completely different places.
Tool Calling Is the Real Test: LLMs Move From Talk to Combat
OpenClaw completely changed how we compare LLMs. It's no longer about who writes the prettiest essay, but who dares to use tools โ and uses them well. Duck Editor's frontline notes.
AI Tech Is Racing Ahead: One Bad News, One Good News
Feeling anxious about how fast AI is developing? Don't panic โ technology will eventually become products anyone can use. The real question is how you position yourself now.
Google's Open-Source Powerhouse Gemma 4: Multimodal AI You Can Run Locally
Google open-sourced Gemma 4 in April 2025 โ multimodal (text, image, audio), runs on consumer hardware, Apache 2.0 licensed. Free for commercial use. Here's what it means.
A Skill Is Really the Evolution of GPTs: Three Stages of Packaging AI Instructions
System prompts, GPTs, Gemini Gems, Skills โ these are essentially the same thing: packaging instructions for AI to use. Duck Editor's take is that they've all evolved in one direction: more convenient, and less for you to worry about.
Dissecting an AI Agent: Skill, Tools, Harness โ Why a Good Model Isn't Enough
The same Claude or GPT, dropped into different AI agents โ one runs as steady as a seasoned chef, another keeps blowing itself up. The difference often isn't the model, but the layer you can't see: the Harness. Duck Editor takes apart the three layers of an agent.
Your Claude Chats Turned Up on Google: The 4th AI 'Share Link' Leak, and the One Lesson to Take Away
In July 2026, a wave of Claude shared conversations got indexed by Google and Bing, exposing API keys, resumes, and crypto wallet data. It's not the first time โ ChatGPT and Grok did it too. The real problem isn't one company slipping; it's a mental-model error: a share link is not a private link.
Computex 2026: Nvidia Breaks the OpenClaw Bottlenecks โ Personal AI Assistant Era Begins
Four months ago I predicted in a course: once the tech matures and good local models arrive, personal AI agents go mainstream โ and that will change the history of personal computing. Computex 2026 just proved it.
Don't Agonize Over the Prompt: Ramble at Your AI by Voice for 10 Minutes Instead
Facing a complex problem? Don't fret over how to word the prompt. Ex-Tesla AI director and OpenAI founding member Karpathy's trick: switch to voice, ramble for ten minutes, and let the AI restructure your mess. The first-principle behind it is exactly what Duck Editor keeps teaching in class โ voice is the lowest-effort way to feed an AI context.