How to Set Up an AI Agent for Your Small Business (No Coding Required)
How to Set Up an AI Agent for Your Small Business (No Coding Required)
Most guides about AI agents assume you’re a developer. This one doesn’t.
If you can type a sentence, you can set up an AI agent that works for your business 24/7. We know because we’ve done it — and the person who set up our first agent had never touched a framework before.
This guide covers everything from installation to your first automated task. Budget about 45 minutes.
What You’ll Need
- A Mac, Linux, or Windows computer
- An AI provider account (we’ll walk you through getting one below)
- About $5–20/month in AI costs depending on usage
- Optional: a Telegram account if you want your agent to message you on your phone
Step 1: Install OpenClaw (5 minutes)
OpenClaw is the open-source framework that turns an AI model into an agent. Open your terminal and paste the install command:
On Mac or Linux:
curl -fsSL https://openclaw.ai/install.sh | bashOn Windows (PowerShell):
iwr -useb https://openclaw.ai/install.ps1 | iexThe installer handles everything automatically and launches the setup wizard when it’s done. That’s the last command you need to paste.
Step 2: Configure Your AI Provider (5 minutes)
The setup wizard runs automatically after installation. It walks you through connecting your AI provider — you’ll pick a provider and paste your API key.
Which model to pick:
- Claude Sonnet — Best starting point. Good at everything, reasonable cost.
- Claude Haiku — Cheaper. Good for simple monitoring tasks.
- GPT-4o — Strong alternative if you prefer OpenAI.
Don’t overthink this. You can change it later.
Step 3: Connect Telegram (5 minutes)
Your agent needs a way to reach you. Telegram is the easiest to set up — you can skip this and add it later, but it’s worth doing now.
- Install Telegram on your phone if you haven’t already (free on App Store / Google Play)
- Open Telegram and search for @BotFather
- Send the message
/newbot - Pick a name and username for your bot
- BotFather gives you a token — copy it and paste it when the OpenClaw wizard asks
Send a message to your new bot. If it replies, you’re connected.
Step 4: Give Your Agent a Memory (15 minutes)
This is the step most people skip — and it’s the reason most people’s AI agents are useless.
Your agent wakes up fresh every session. It has no idea who it is, who you are, or what it’s supposed to be doing. This step fixes that — and you don’t have to create any files yourself. Just talk to your agent.
Open a conversation with your agent and say these things:
Tell it who you are:
“I want you to remember who I am between sessions. Set up your memory files. My name is [your name], I run [your business], and I’m in [your timezone].”
Give it a personality:
“I want you to have a real personality. Be direct, honest, and proactive. Don’t ask obvious questions — figure things out and come to me when you’re actually stuck.”
Set up a status tracker:
“Set up a way to track what we’re working on. I want to be able to see active projects, any blockers, and what you did today — without having to ask.”
Tell it how to start each session:
“Every time you start a session, read your memory files first so you know who I am and what we’ve been working on. Don’t start fresh — start informed.”
Start a lessons-learned file:
“Keep a running file of mistakes and lessons. Any time something goes wrong or we figure out a better way to do something, write it down. I want you learning from experience, not repeating the same errors.”
Your agent will build out the files it needs based on what you’ve told it. The result is the same as manually creating five workspace files — except you described what you wanted and it handled the structure.
Why this matters: These files are loaded every time your agent starts a session. They’re the difference between talking to a stranger and talking to someone who knows you, knows the project, and knows what happened yesterday.
Step 5: Set Up Your First Heartbeat (5 minutes)
A heartbeat is a periodic check-in. Your agent wakes up, reviews what needs attention, and either does work or stays quiet.
Tell your agent:
“Check in every hour. Look at what we’re working on, see if anything needs my attention, and if something does — message me on Telegram. If nothing needs attention, stay quiet.”
That’s it. Your agent now has a rhythm. It’ll read your workspace files each hour, see if anything needs attention, and act accordingly — without you having to prod it.
Step 6: Create Your First Scheduled Task (5 minutes)
Let’s make sure you start every morning knowing what’s on your plate.
Tell your agent:
“Every morning at 9 AM, check what’s happening — active projects, blockers, anything coming up today — and send me a summary on Telegram. Keep it to 3 bullet points. If everything’s clear, just say ‘All clear.’”
Tomorrow at 9 AM, you’ll get a Telegram message with your daily snapshot. Set it and forget it.
Step 7: Give It Something to Do
Now that your agent has memory, a heartbeat, and a morning check-in, give it real work.
Tell your agent:
“Research 5 competitors in [your niche] and summarize their pricing, positioning, and what seems to be working for them. Save it somewhere I can find it later.”
On the next heartbeat, your agent will pick this up and start. By the time you check in, it’ll have something for you.
From here, you can keep giving it work the same way — just tell it what you want done. It’ll track the task, work on it, and report back.
What Happens Next
Over the next few days, you’ll notice:
- Your agent starts building daily logs of what it worked on
- Its lessons-learned file gets its first entry (this is good — it means the system is learning)
- You’ll develop a natural rhythm of briefing it in the morning and reviewing in the evening
- The heartbeat becomes the rhythm of your operation
Day 2 priorities:
- Tell your agent about a real ongoing project and let it work overnight
- Ask it to suggest what else it should be tracking
- Set up a second scheduled task (weekly review, content ideas, etc.)
Troubleshooting {#troubleshooting}
Agent doesn’t respond on Telegram: Open a chat with your agent through the dashboard and ask it: “Check if my Telegram connection is working and fix it if it’s not.”
Responses are generic/unhelpful: Your agent has no context about you or your business. Go back to Step 4 and have the conversation. The more specific you are about who you are and what you do, the better it performs.
Heartbeat isn’t firing: Tell your agent: “Check if my heartbeat is configured correctly and if the gateway is running as a background service. Fix anything that’s wrong.”
Something else not working? Tell your agent what’s happening. Seriously — that’s the whole point. Describe the problem and let it troubleshoot. If it can’t fix it, check the OpenClaw docs or the community Discord.
For the complete guide to building an AI-operated business, see The OpenClaw Playbook.