AI Realist Live Workshop · May 2026

Build an AI Agent
From Scratch

A hands-on live workshop for AI Realist subscribers. Build, deploy, and share your own OpenClaw agent — no coding background needed.

May 26, 2026 · 18:00–20:00 CET 2 hours · live & interactive For AI Realist subscribers

Want to join the workshop? Become a paid subscriber of AI Realist — any paid tier gives you access.

Subscribe now →
What you'll build

Your First OpenClaw Agent

Create, configure, and deploy your own AI agent using OpenClaw. You'll build a fully functional agent that can perform real tasks — and publish it to the web with Cloudflare Pages. No coding background needed.

🔑 You'll need an API key

To follow the workshop, you need an API key for an AI model provider. The cheapest option is DeepSeek directly (platform.deepseek.com) — pennies per session. But you can use OpenAI, Anthropic (Claude), OpenRouter, or any other provider OpenClaw supports. Step 02 below covers how to configure your key — pick whichever provider works best for you.

Before the workshop
⚠
Required — please complete before May 26

We won't have time to troubleshoot installations during the session. This takes roughly 10 minutes. Complete it in advance so you can start building from minute one.

✨

First time using Terminal?

Terminal (Mac) and PowerShell (Windows) are just text-based ways to give your computer instructions. Instead of clicking icons, you type commands and press Enter. Don't worry — we'll tell you exactly what to type.

On a Mac

  1. 1. Press ⌘ Command + Space — this opens Spotlight Search
  2. 2. Type Terminal and press Return (Enter)
  3. 3. A dark window appears — that's Terminal! ✅

On Windows

  1. 1. Press Win + R — this opens the Run dialog
  2. 2. Type powershell and press Enter
  3. 3. A blue window appears — that's PowerShell! ✅
How to use the commands on this page

1. Click the "Copy" button next to any command — it copies the text automatically.
2. Go to Terminal / PowerShell and right-click → Paste (or press Ctrl+V on Windows / ⌘+V on Mac).
3. Press Enter / Return — the command runs and shows you the result.

💡 What do these commands do?

Every box with a dark background is a command — an instruction you give to your computer. We've added short explanations underneath each one so you always know what's happening. If something goes wrong, the error message usually tells you what's missing.

00

Install Python 3

The scanner we'll build runs on Python 3. Most computers already have it — check once, and if it's there, skip this step entirely.

  1. 1

    Check if Python is already installed:

    Terminal · macOS 
    python3 --version

    If you see Python 3.x.x — you're done.

  2. 2

    If not installed — go to python.org/downloads and click the big yellow Download Python 3.x.x button. Open the downloaded file and click through the installer.

  1. 1

    Check if Python is already installed:

    PowerShell · Windows 
    python --version

    If you see Python 3.x.x — you're done.

  2. 2

    If not installed — the fastest way is one command in PowerShell:

    PowerShell · Windows 
    winget install Python.Python.3

    Or go to python.org/downloads, download the installer, and run it. Important: on the first installer screen, tick "Add Python to PATH" before clicking Install.

01

Install OpenClaw

OpenClaw is the tool we'll use throughout the session. Select your OS below and run the command in your terminal.

docs.openclaw.ai/install
  1. 1

    Open Terminal — see how below if you're not sure

  2. 2

    Paste and run this command:

    Terminal · macOS 
    curl -fsSL https://openclaw.ai/install.sh | bash

    Official installer — detects your OS, installs Node if needed, installs OpenClaw, and launches the setup wizard.

  3. 3

    Follow the on-screen prompts to finish the wizard.

How to open Terminal macOS

  1. Press ⌘ Command + Space to open Spotlight Search
  2. Type Terminal and press Return
  3. Click the Copy button on the command above — then right-click → Paste into Terminal (or press ⌘+V)
  4. Press Return to run it
Using an AI coding assistant?

If you have Claude Code, GitHub Copilot, or GitHub Codex — paste the install command directly and ask it to guide you through setup. Say: "Help me run this OpenClaw install command."

  1. 1

    Open PowerShell — see how below

  2. 2

    Paste and run this command:

    PowerShell · Windows 
    & ([scriptblock]::Create((iwr -useb https://openclaw.ai/install.ps1)))

    Official Windows installer. PowerShell works well; WSL2 also supported for extra stability.

  3. 3

    Follow the on-screen prompts to finish the wizard.

⚠ If you get a "SecurityException" error

Windows blocks scripts by default. If you see an error like:
SecurityException : The term 'SecurityException' is not recognized...
You're hitting a PowerShell execution policy block. Run this command first to allow scripts, then retry the install:

PowerShell · Windows 
Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

If that still fails, use this temporary bypass instead:

PowerShell · Windows 
Set-ExecutionPolicy -ExecutionPolicy Bypass -Scope Process

How to open PowerShell Windows

  1. Press Win + R to open the Run dialog
  2. Type powershell and press Enter
  3. Click the Copy button on the command above — then right-click → Paste into PowerShell (or press Ctrl+V)
  4. Press Enter to run it
Using an AI coding assistant?

If you have Claude Code, GitHub Copilot, or GitHub Codex — paste the install command directly and ask it to guide you through setup. Say: "Help me run this OpenClaw install command."

📺 During setup — you can skip everything for now

The wizard will ask about channels (Telegram, WhatsApp, Signal, etc.), skills, and other extras. You can safely skip everything for now — just choose "Skip for now" on each screen. For the workshop, we'll be using the built-in Web UI, which doesn't need any of this. Everything can be added later.

When asked "How do you want to hatch your agent?", choose "Hatch in Browser". This opens the chat in your web browser — exactly what we need for the workshop.

02

Configure OpenRouter

OpenClaw uses OpenRouter to access AI models. Connect your API key — most models have free tiers, paid ones cost pennies per session.

docs.openclaw.ai — OpenRouter provider openrouter.ai — OpenClaw integration guide
Easiest: use the guided wizard

The onboard command walks you through the entire OpenRouter setup interactively. Recommended if this is your first time.

Terminal · macOS & Windows 
openclaw onboard

When you see the prompt ◆ Model/auth provider, choose More, then select OpenRouter from the list and press Enter. Paste your OpenRouter API key (begins with sk-or-...). If you don't have one, sign up at openrouter.ai/keys.

⚡ Fastest: one command

Already have your OpenRouter key? Run this single command — replace the key with yours:

Terminal · macOS & Windows 
openclaw onboard --auth-choice apiKey --token-provider openrouter --token "sk-or-v1-your-key-here"

Replace sk-or-v1-your-key-here with your actual key.

Or set the key directly without the wizard:

  1. 1

    Open Terminal (same as Step 1)

  2. 2

    Set your OpenRouter key:

    Terminal · macOS 
    openclaw config set models.providers.openrouter.apiKey "YOUR_OPENROUTER_KEY"

    Replace YOUR_OPENROUTER_KEY with your key from openrouter.ai/keys (begins with sk-or-...).

  3. 3

    Verify the key was saved:

    Terminal · macOS 
    openclaw models auth list

    You should see openrouter:default listed as a configured provider.

Or set the key directly without the wizard:

  1. 1

    Open PowerShell (same as Step 1)

  2. 2

    Set your OpenRouter key:

    PowerShell · Windows 
    openclaw config set models.providers.openrouter.apiKey "YOUR_OPENROUTER_KEY"

    Replace YOUR_OPENROUTER_KEY with your key from openrouter.ai/keys (begins with sk-or-...).

  3. 3

    Verify the key was saved:

    PowerShell · Windows 
    openclaw models auth list

    You should see openrouter:default listed as a configured provider.

03

Launch & Verify

Start OpenClaw and confirm your agent responds to your messages.

Terminal / PowerShell 
openclaw dashboard

This opens the Web UI at http://127.0.0.1:18789. Send a test message: "Hello! What model are you running?"

If your agent responds — your setup is complete! ✅

No response in the chat?

Run the doctor:

Terminal / PowerShell 
openclaw doctor --fix

Still stuck? Restart and retry:

Terminal / PowerShell 
openclaw doctor --fix && openclaw gateway restart
🔁 Device pairing required?

If you see "Device pairing required" in the browser, copy the device code and run:

openclaw devices approve YOUR-DEVICE-CODE

Then reconnect in the browser.

No response in the chat?

Run the doctor command — it detects common issues and can fix them automatically:

Terminal · macOS & Windows 
openclaw doctor --fix

Also check that openclaw models auth list shows openrouter:default, and that the model name is spelled exactly as shown above — including the openrouter/ prefix.

🔄 If any error persists after fixing

Restart OpenClaw and try the doctor — most issues clear up after a restart:

Terminal · macOS & Windows 
openclaw gateway restart

If the problem still doesn't go away, run the doctor with auto-fix and restart again:

Terminal · macOS & Windows 
openclaw doctor --fix && openclaw gateway restart
04

Prepare Cloudflare Pages

We'll be deploying to Cloudflare Pages during the workshop. Get your account ready beforehand.

Create a Cloudflare account

Go to dash.cloudflare.com/sign-up and create a free account. No credit card required — the Pages free tier is genuinely free.

Install Wrangler CLI

Wrangler is Cloudflare's CLI for deploying to Pages. Node.js was installed automatically with OpenClaw, so npm is already available.

Terminal / PowerShell 
npm install -g wrangler

Authenticate

Terminal / PowerShell 
npx wrangler login

Your browser opens to authenticate with Cloudflare. Verify it worked:

Terminal / PowerShell 
npx wrangler whoami

If it shows your Cloudflare account email — you're authenticated and ready for the workshop.

🦞 Already have Wrangler?

Run npx wrangler --version to check. If you see a version, just make sure you're logged in with npx wrangler login.

🚀
One final thing: launch & verify

Before wrapping up, make sure your agent responds in the Web UI:

openclaw dashboard

Send a test message: "Hello! What model are you running?"

If your agent responds — your setup is complete! ✅

Pre-workshop checklist

Confirm all six before Tuesday, May 26:

☑ Become a paid AI Realist subscriber on Substack
☑ Python 3 installed — python3 --version (macOS) or python --version (Windows) prints a version number
☑ OpenClaw installed — openclaw --version prints a version number
☑ OpenRouter key set — openclaw models auth list shows openrouter:default
☑ Agent responds in the Web UI — you sent a test message and got a reply
☑ Cloudflare account + Wrangler authenticated — npx wrangler whoami shows your email
Need help before the session?

Check the OpenClaw FAQ for common installation issues.

💬 Still stuck? Leave a comment on the AI Realist Live Workshop post — I'll help you out.

⚠️ Stuck on opening Terminal? See the ✨ First time using Terminal? section above for step-by-step instructions.

Tips & Tricks

Getting the Most from OpenClaw

Practical advice — model selection, performance tuning, and routing strategies.

⚡ Use DeepSeek V4 Directly

Configure DeepSeek as a direct provider — faster, cheaper, and fewer rate limits than routing through OpenRouter.

Configuration

Terminal 
openclaw config set models.providers.deepseek.apiKey "sk-your-key-here"
openclaw config set models.providers.deepseek.baseUrl "https://api.deepseek.com/v1"

Usage

Terminal 
# Use the model directly (no routing)
openclaw chat --model deepseek/deepseek-v4-flash

# In gateway/webchat — prefix with provider name:
# deepseek/deepseek-chat
# deepseek/deepseek-reasoner
⚡ Why direct

Lower latency (no routing hop), fewer outages (independent of OpenRouter), full context window, and support for token tracking.

🧠 Switch Off Thinking

Thinking mode is great for complex reasoning but wasteful for simple tasks. Control it per-session.

Terminal 
# In TUI / CLI:
openclaw chat --thinking off

# In isolated tasks (cron):
# Set payload.thinking to "disabled" or omit it
ModeBest ForAvoid When
On Complex reasoning, debugging, multi-step analysis, quality review Simple Q&A, formatting, boilerplate tasks
Off Fast Q&A, file operations, routine cron tasks Deep logical reasoning, multi-step validation
💡 Rule of thumb

If the task can be done in one or two sentences — thinking off. If it needs analysis, comparison, or planning — keep it on.

🔀 Use Models Directly from Provider

OpenRouter is great for exploration. For production — go direct to the provider.

Example configs 
# DeepSeek
openclaw config set models.providers.deepseek.apiKey "sk-..."
openclaw config set models.providers.deepseek.baseUrl "https://api.deepseek.com/v1"

# Google / Gemini
openclaw config set models.providers.google.apiKey "..."

# OpenAI
openclaw config set models.providers.openai.apiKey "sk-..."

# Anthropic / Claude
openclaw config set models.providers.anthropic.apiKey "sk-ant-..."
AspectOpenRouterDirect Provider
Latency+50–200ms routing overheadDirect, no hop
Rate limitsShared poolYour own quota
Provider featuresLimited (no streaming metadata, limited tool-calling)Full API surface
PricingSmall markupDirect rates
Discovery✅ One key for many models❌ One key per provider
🎯 Best practice

Use OpenRouter for exploration (testing models, comparing outputs) and direct providers for production (daily workflow, cron jobs, agents). Configure both and switch with --model provider/model-name.

⚙️ Default Models & Fallbacks

Configure OpenClaw to use specific models by default and fall back gracefully when a provider is down.

Terminal 
# Set the default model for all agents
openclaw config set agents.defaults.model "deepseek/deepseek-v4-flash"

# Configure fallback chain
openclaw config set agents.defaults.fallbacks   '["deepseek/deepseek-chat", "openai/gpt-4o"]'

# Force per-agent override
openclaw agents update your-agent   --model deepseek/deepseek-v4-pro   --thinking high