> ## Documentation Index
> Fetch the complete documentation index at: https://docs.radicalwhale.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Creating Agents

> Create and configure AI agents

## Creating an Agent

Click **Agents** in the sidebar → **Create Agent**.

**Agent Name**: Descriptive name like "Company Enrichment Agent" or "Resume Parser". Avoid generic names.

**Description**: Explain what the agent does, what data it processes, and expected outputs.

```
Enriches company records by analyzing names and websites to extract
industry classification, company size, headquarters location, and
business focus areas using web search and company databases.
```

## Configure AI Model

**GPT-4**: Best for complex reasoning and detailed analysis. Higher cost, slower.

**GPT-4 Mini**: Balanced performance and cost. Good for standard data processing.

**GPT-4 Nano**: Fast, low cost, simple tasks only.

**API Key**: Select from existing variables or create new. Never paste keys directly—always use secure variables.

## Write System Prompt

The system prompt defines your agent's role, instructions, output format, and guidelines.

### Structure

**Role**: Define who the agent is

```
You are a professional business research analyst specializing in
company intelligence and market research.
```

**Task**: Explain what to do

```
When provided with company information (name, website, description):
1. Research and analyze the company's core business
2. Determine primary industry and market sector
3. Estimate company size and maturity level
4. Identify key products or services
5. Assess market position
```

**Output Format**: Specify exact format

```
Provide analysis in JSON format:
{
  "industry": "Primary industry classification",
  "sector": "Market sector (B2B, B2C, etc.)",
  "company_size": "Startup | Small | Medium | Large | Enterprise",
  "primary_products": ["List of main products/services"],
  "confidence_score": 0.85
}
```

**Guidelines**: Add important rules

```
- Always be factual and cite sources when possible
- If information is uncertain, indicate confidence level
- Focus on publicly available information
- If cannot find sufficient information, clearly state this
```

## Select Tools (Optional)

Choose tools that extend agent capabilities: web search, company databases, social media APIs, document processing, custom APIs.

<Info>
  Start with fewer tools and add as needed. Too many tools slow down agents and
  increase costs.
</Info>

## Test Your Agent

Click **Test Agent** → Provide sample input → Review response → Check formatting, accuracy, completeness.

Test with typical cases, edge cases, missing information, and ambiguous data.

## Save and Deploy

Review configuration → **Create Agent** → Agent is now active.

## Connect to Dataset

Navigate to dataset → Click column header → **Edit Column** → Select your agent in Agent dropdown → Add column-specific instructions if needed → Save.

## Next Steps

<CardGroup cols={2}>
  <Card title="Test Agents" icon="flask" href="/guides/agents/testing-agents">
    Test and validate agents
  </Card>

  <Card title="Chat with Agents" icon="message" href="/guides/agents/chatting-with-agents">
    Interactive agent testing
  </Card>

  <Card title="Manage Agent Tools" icon="wrench" href="/guides/agents/managing-tools">
    Add and configure tools
  </Card>

  <Card title="Configure Models" icon="brain" href="/guides/agents/configuring-models">
    Advanced model configuration
  </Card>
</CardGroup>