Zero to Agent in Under an Hour
Everything you need to set up your first AI agent for your business. No coding required.
In This Guide
What You'll Need
- A Mac, Linux, or Windows computer
- A ChatGPT account or an API key from an AI provider (we'll walk you through the options below)
- About $5–20/month in AI costs depending on how much you use it
- Optional: a Telegram account if you want your agent to message you on your phone
Step 1: Install OpenClaw (2 minutes)
Open your terminal and run the installer. It handles everything automatically — detects your system, installs what's needed, and launches the setup wizard.
On Mac or Linux, paste this and press Enter:
curl -fsSL https://openclaw.ai/install.sh | bash On Windows, open PowerShell and paste this:
iwr -useb https://openclaw.ai/install.ps1 | iex That's the last time you need to paste a command. The installer launches the onboarding wizard automatically.
Step 2: Walk Through the Setup Wizard (5 minutes)
After installation, the onboarding wizard runs automatically. It walks you through each choice with clear prompts — no guessing required. Here's what to expect and how to choose:
Choose where to run your Gateway
The Gateway is the engine that keeps your agent running. Think of it like a home base — it's the program on your computer that your AI agent "lives" in.
The wizard asks where to run it:
- This computer (Local) — Pick this if you're just getting started. Your agent runs on your machine. Simple, no extra setup.
- Remote server — For later, when you want your agent running 24/7 on a server you can walk away from.
Our recommendation: Start local. You can always move to a server later.
Pick your AI provider
The wizard shows you a list of AI providers. Depending on the option you choose, you'll either sign in with OAuth or paste an API key.
Don't overthink the provider choice — you can change it later or even use multiple at once.
Pick a model
After choosing a provider, you'll pick a specific model:
- GPT-5.4 — Recommended starter. Cost-efficient and strong enough for most daily agent, coding, and operations work.
- Claude Opus — Best reasoning quality for deeper orchestration and architecture work, but more expensive per turn.
- Claude Haiku — Faster and cheaper for lightweight monitoring, summaries, and check-ins.
Starter (recommended): GPT-5.4 via OpenAI Codex OAuth. Cost-efficient. Sign in with your ChatGPT account in OpenClaw.
Pro (for deeper orchestration): Claude Opus. Best reasoning quality. More expensive per turn, so upgrade when you start hitting limits on strategic or architectural work.
GPT-5.4 runs via OpenAI Codex. Codex is currently included with ChatGPT Free, Go, and all paid tiers; paid tiers get higher rate limits. API key auth is also supported for pay-as-you-go usage.
Connect a messaging channel (optional)
This step lets your agent message you on your phone — like having a team member who can text you when something needs attention. You can skip this and add it later, but Telegram is the easiest to start with.
Other supported channels include WhatsApp, Discord, iMessage, Slack, and more. You can add any of them later.
Install as a background service (recommended)
The wizard asks if you want the Gateway to start automatically. Say yes.
Step 3: Verify Everything Works (1 minute)
After onboarding finishes, it opens a chat window where you can talk to your agent immediately. If your agent responds to a message, everything is working.
You can also open the dashboard anytime:
openclaw dashboard This opens a web interface in your browser where you can chat, see your agent's status, and manage settings.
Step 4: Teach Your Agent Who You Are (15 minutes)
This is the step most people skip — and why most AI agents feel useless. Without context, your agent is just a chatbot. With context, it becomes a team member that knows your business.
Open the chat with your agent (through the dashboard or Telegram) and introduce yourself. Here's what to say:
Say this to your agent:
"I want you to set up your workspace so you remember who I am between sessions. Create the files you need. Here's what to know about me: My name is [your name]. I'm in [your timezone]. I run [describe your business or what you do]. I prefer [how you like to communicate — brief updates? detailed reports? casual tone?]. My current focus is [your main project or goal right now].
Also set up a personality for yourself — be genuinely helpful, have opinions, figure things out before asking me, and surface blockers immediately instead of sitting on problems."
Your agent will create the files it needs to remember you — your preferences, your business context, and how you like to work. You don't need to write or organize any of this yourself.
Step 5: Set Up Automatic Check-Ins (5 minutes)
This is where your agent stops being a chatbot and starts being an operator. A "heartbeat" is an automatic check-in — your agent wakes up on a schedule (say, every hour), looks at your projects and tasks, and either takes action or stays quiet if nothing needs attention.
You don't have to be in a conversation for this to happen. It runs on its own, like a team member checking in throughout the day.
Say this to your agent:
"Set up an automatic heartbeat that runs every hour. When it fires, check if there are any blockers, upcoming deadlines, or tasks that need attention. If something's urgent, message me on Telegram. If everything's fine, just note that you checked and move on."
Your agent handles the configuration. You'll start getting proactive updates without asking for them — that's the difference between a chatbot and an operator.
Step 6: Give It Something Real to Do (5 minutes)
The best way to test your agent is with a real task, not a toy example. Here are some good first assignments:
Morning briefing:
"Set up a scheduled task that runs every morning at 9 AM. Check my current projects for any blockers or upcoming deadlines, and send me a 2-3 bullet summary on Telegram. If nothing needs attention, just say 'all clear.'"
Research task:
"Research [topic relevant to your business] and write up a summary of what you find. Save it somewhere I can review it later."
Content help:
"Draft a blog post about [topic]. I want it to be [tone/length/audience]. Save it as a draft and let me know when it's ready for review."
Start with one task. When you see it deliver, you'll start thinking of ten more.
What's Next
Over the next few days, you'll notice your agent building daily logs, learning from mistakes, and getting better at anticipating what you need.
Day 2 priorities:
- Tell your agent about a real project and let it work on it
- Set up a second scheduled task (weekly review, competitor monitoring, etc.)
- Let it run overnight and check the daily log in the morning
The key insight: the skill you're building is how to communicate with your agent effectively. The better you describe what you want, the better the results. That's a skill that transfers to every AI tool, not just OpenClaw.
Want the full guide?
The OpenClaw Playbook covers advanced workflows, multi-agent patterns, content automation, and everything that happens after day one.
Get the Playbook →