How to Create an AI Agent with OpenClaw (No Code Required)

Why Create an AI Agent?

AI agents are autonomous programs powered by large language models that can perform real work on your behalf. Unlike chatbots that only answer questions, agents can research topics, write blog posts, analyze SEO data, draft email campaigns, manage customer support tickets, and coordinate with other agents — all without constant human supervision.

For solopreneurs and small teams, AI agents replace the need to hire specialists for every function. A single content writing agent can produce blog posts, social media copy, and email sequences at a fraction of the cost and time. A research agent can monitor competitors and surface insights daily. An SEO agent can audit your site and recommend optimizations continuously.

OpenClaw is the easiest way to create these agents. It uses a markdown-based configuration system called SOUL.md that requires zero coding knowledge. You describe what your agent should do in plain English, pick a language model, and start it up. The rest of this guide shows you exactly how.

Prerequisites

Before you begin, make sure you have the following ready. The setup is minimal — most developers and non-technical users already have everything they need.

Once you have these three things, you can install OpenClaw with a single command:

npm install -g openclaw

Step 1: Initialize Your Project

OpenClaw provides an init command that scaffolds everything you need. Open your terminal, navigate to the folder where you want your agent project to live, and run:

npx openclaw init my-agent

This command creates a new directory called my-agent with the following structure:

my-agent/
├── agents/
│   └── default/
│       └── SOUL.md      ← Your agent's identity
├── knowledge/           ← Shared files your agent can read
├── agents.md            ← Agent registry (for multi-agent)
└── openclaw.config.json ← Project settings

The most important file is agents/default/SOUL.md. This is where you define who your agent is and what it does. OpenClaw generates a starter SOUL.md with placeholder content — your next step is to replace it with your own configuration.

Step 2: Configure Your Agent with SOUL.md

SOUL.md is the heart of your agent. It is a plain markdown file that defines your agent's role, personality, rules, output format, and available tools. The language model reads this file before every interaction, so everything you write here shapes how your agent behaves.

Here is a complete SOUL.md example for a Content Writer agent. Open agents/default/SOUL.md in any text editor and replace the placeholder content:

# ContentWriter

## Role
You are a content marketing specialist. You write
blog posts, email campaigns, social media copy, and
landing page content for SaaS companies.

## Personality
- Tone: Professional but approachable
- Style: Clear, concise, scannable paragraphs
- Voice: Active voice, short sentences
- Always support claims with data or examples
- Use analogies to explain technical concepts

## Rules
- ALWAYS respond in English
- Target 1,200-1,800 words for blog posts
- Include a meta description (max 155 characters)
- Never use clickbait or misleading titles
- Every blog post needs an intro, 3-5 H2 sections,
  and a conclusion with a call to action
- Do NOT use emojis in written content

## Output Format
- Title (H1)
- Meta description
- Body with H2 section headers
- Conclusion with CTA
- 3-5 suggested tags

## Tools — USE THEM
- Use Browser to research topics and check competitor
  content before writing
- Use Search to find relevant data and statistics
- Use File operations to save drafts to the
  knowledge base

Let's break down what each section does. The Role section tells the agent what it is and what domain it operates in — this is the most important section because it anchors all of the agent's behavior. The Personality section controls tone and communication style, ensuring consistent output that matches your brand. The Rules section sets hard constraints using strong language like ALWAYS and NEVER — these are your guardrails. The Output Format section defines the structure of the agent's deliverables so you get consistent results every time. And the Tools section tells the agent which capabilities to use and when — without this, agents often ignore their available tools even when they would be helpful.

The key principle is specificity. Vague instructions like "write good content" produce inconsistent results. Specific instructions like "target 1,200-1,800 words with H2 headers every 200-300 words" produce reliable, high-quality output every time.

Step 3: Set Your Model

Your agent needs a language model to power its reasoning. OpenClaw supports all major model providers. Choose the one that fits your use case and authenticate with your API key.

For Anthropic Claude (recommended for writing and reasoning tasks):

openclaw models auth setup-token \
  --provider anthropic

For OpenAI GPT-4o (good for general-purpose and code tasks):

openclaw models auth setup-token \
  --provider openai

The command will prompt you to paste your API key. Once authenticated, you can configure which specific model your agent uses in the openclaw.config.json file. Claude Sonnet is a great default for most tasks — it balances quality, speed, and cost. Use Claude Opus for complex reasoning tasks or GPT-4o for code-heavy workflows. You can change models at any time without modifying your SOUL.md.

Step 4: Add Skills (Tools)

Skills are the actions your agent can take beyond text generation. By default, OpenClaw agents can read and write files. But you can enable additional skills to give your agent more capabilities.

Skills are enabled in your project configuration and referenced in your SOUL.md. The key is to not just enable a skill — you must also tell your agent when to use it. An agent with Browser access but no instructions about when to browse will rarely use it. Add explicit instructions like "Use Browser to research the topic before writing any content" in your SOUL.md's Tools section.

Step 5: Start Your Agent

Everything is configured. Now start the OpenClaw gateway to bring your agent online:

openclaw gateway start

The gateway starts on port 18789 by default. You will see a confirmation message with your agent's name and the model it is using. Your agent is now ready to receive tasks.

Open a new terminal window and send your first message:

openclaw chat "Write a 1,500-word blog post about
the benefits of AI automation for small businesses"

Your agent reads its SOUL.md, understands its role as a content writer, follows the rules you defined, uses the output format you specified, and delivers the blog post. If you enabled the Browser skill and instructed the agent to research before writing, it will visit relevant pages first and incorporate findings into the post. That is the power of a well-configured SOUL.md — the agent does exactly what you told it to do, consistently, every time.

Creating Multiple Agents

Once your first agent is working, you can add more agents to build a full team. Each agent lives in its own directory under agents/ with its own SOUL.md file. The agents.md file at the project root registers all agents and defines how they relate to each other.

my-agent/
├── agents/
│   ├── content-writer/
│   │   └── SOUL.md
│   ├── seo-analyst/
│   │   └── SOUL.md
│   └── researcher/
│       └── SOUL.md
├── knowledge/
├── agents.md              ← Registers all agents
└── openclaw.config.json

In each agent's SOUL.md, you can add a Handoffs section that tells the agent when to pass work to another agent. For example, your researcher might hand off to @ContentWriter when research is complete, and your content writer might hand off to @SEOAnalyst when the draft is ready for optimization. This creates an automated pipeline where agents collaborate without manual intervention.

For advanced multi-agent orchestration — task routing, parallel execution, shared dashboards, and cross-framework agent coordination — read our multi-agent system guide. CrewClaw adds an orchestration layer on top of OpenClaw that lets you manage agent crews from a single dashboard, monitor agent activity in real time, and define complex workflows visually.

Use the SOUL.md Generator (No Terminal)

If you prefer not to use the terminal at all, CrewClaw offers a free SOUL.md Generator that lets you create agent configurations through a visual form. The generator includes 47 pre-built templates for common agent roles — content writer, SEO analyst, research specialist, project manager, data analyst, customer support, and many more.

The process is straightforward: select a template or start from blank, fill in your agent's name, role, personality traits, rules, and tool preferences using the form fields, then click download to get a ready-to-use SOUL.md file. Drop that file into your OpenClaw project's agent directory and your agent is configured.

The generator is especially useful for non-technical users, teams who want standardized agent configurations, and anyone who wants to quickly prototype different agent roles before committing to a setup. Every template is based on real-world agent configurations that have been tested and optimized.

Frequently Asked Questions