How to Set Up Your First AI Agent with OpenClaw
Setting up your first AI agent is the moment everything clicks — you go from "I installed OpenClaw" to "I have an AI employee working for me." This guide covers agent configuration from start to finish: writing effective Soul.md instructions, choosing the right AI model, connecting communication channels, and tuning your agent until it performs reliably. Follow these steps and you will have a working AI agent in about an hour.
What You Need
- ✓OpenClaw installed and running on your server
- ✓At least one AI provider API key configured
- ✓Clear idea of what you want the agent to do
- ✓About 60-90 minutes for thorough configuration
Step-by-Step Guide
Define your agent's purpose
Before touching any settings, write down exactly what you want your agent to do. Be specific: "Handle customer inquiries about our products via WhatsApp and email, qualify leads, and schedule consultations." A clear purpose leads to clear instructions and better agent performance. Avoid trying to do everything at once — start with one primary responsibility and expand later.
Write your Soul.md
Soul.md is your agent's brain. Start with: (1) Identity — who is this agent? Give it a name and role. (2) Personality — how should it communicate? Professional, casual, friendly? (3) Knowledge — what does it know about your business? Products, services, pricing, policies. (4) Rules — what should it always or never do? (5) Escalation — when should it hand off to a human? Write in clear, natural language. The AI interprets these instructions, so be specific and unambiguous.
Choose your AI model
OpenClaw supports 35+ AI models. For most business agents, we recommend starting with Claude Sonnet or GPT-4o — they offer the best balance of intelligence and cost. For simple FAQ-style responses, a smaller model reduces costs. For nuanced customer interactions, a more powerful model provides better results. You can change models at any time and even use different models for different tasks.
Connect communication channels
Add the channels your customers use — WhatsApp, Telegram, email, or a combination. Each channel has its own setup process (covered in our dedicated integration guides). Start with one channel, verify it works perfectly, then add more. Multi-channel setups work seamlessly — the agent maintains context regardless of which channel the customer uses.
Set up scheduling (cron jobs)
Configure the tasks your agent should perform on a schedule. Common cron jobs include: checking email every 5-10 minutes, sending daily summaries, and performing weekly follow-ups. Start conservative — you can always add more scheduled tasks as you get comfortable with the system.
Test rigorously
Send your agent at least 20-30 diverse test messages covering different scenarios. Include easy questions, tricky questions, off-topic messages, and edge cases. Review every response and adjust Soul.md instructions where needed. This testing phase is critical — invest the time here to avoid surprises when real customers interact with your agent.
Common Mistakes to Avoid
- !Writing Soul.md instructions that are too general — "be helpful" is not enough, specify exactly what being helpful looks like
- !Choosing a model that is too powerful (expensive) for simple tasks
- !Skipping the testing phase — always test before going live
- !Not planning for edge cases — what should the agent do when it does not know the answer?
- !Trying to configure everything at once — start simple and iterate
Want the full walkthrough? This guide covers the essentials, but the CampeloClaw course provides detailed video instruction for every step, troubleshooting guides, and hands-on practice exercises.
Frequently Asked Questions
How long should my Soul.md be?
Most effective Soul.md files are 500-3,000 words. Enough to be comprehensive but concise enough that the AI processes it efficiently. Quality of instructions matters more than quantity.
Can I change the AI model later?
Yes. You can switch AI models at any time through the OpenClaw dashboard. Your agent's configuration and memory remain intact. Many users experiment with different models to find the best fit.
Related Pages
Master OpenClaw — From Zero to 24/7 AI Assistant
Learn everything in this guide and more with step-by-step video lessons, hands-on projects, and lifetime updates. Join hundreds of students already building their AI workforce.
Get Full Course Access →