Agents

Creating Intelligent Agents

Agents are the fundamental building blocks of your AI-powered automation. Each agent is a specialized AI entity with specific capabilities, knowledge, and behaviors designed to accomplish particular tasks.

What is an Agent?

An agent is an autonomous AI entity that can understand context, make decisions, use tools, and complete tasks. Unlike simple chatbots, agents have:

Intelligence

Tool Access

Can use tools to search the web, query databases, send emails, generate images, and interact with external APIs.

Memory

Short-term context within conversations and long-term memory for learning from past interactions.

Guardrails

Built-in safety constraints, validation rules, and approval gates to ensure responsible AI behavior.

Agent Types

RealAroha supports several agent types, each optimized for different use cases:

Assistant Agent

Most Common

General-purpose conversational agent for interactive tasks

{
  "agentType": "assistant",
  "capabilities": {
    "canUseTools": true,
    "canTriggerApprovals": true,
    "canCallOtherAgents": false,
    "canExecuteCode": false
  }
}

Use cases: Customer support, content creation, data analysis, research assistance

Coordinator Agent

Multi-Agent

Orchestrates multiple agents and manages complex workflows

{
  "agentType": "coordinator",
  "capabilities": {
    "canUseTools": true,
    "canTriggerApprovals": true,
    "canCallOtherAgents": true,  // Key capability
    "canExecuteCode": false
  },
  "delegationRules": {
    "allowedAgents": ["content-writer", "researcher", "reviewer"],
    "maxDelegationDepth": 3
  }
}

Use cases: Project management, complex workflows, multi-step campaigns

Specialist Agent

Highly focused agent for specific domain tasks

{
  "agentType": "specialist",
  "specialization": "content-writing",
  "capabilities": {
    "canUseTools": true,
    "canTriggerApprovals": false,
    "canCallOtherAgents": false,
    "canExecuteCode": false
  },
  "expertiseAreas": ["copywriting", "SEO", "social-media"]
}

Use cases: Content writing, image generation, data transformation, code review

Scheduled Agent

Runs on triggers or schedules without human initiation

{
  "agentType": "scheduled",
  "capabilities": {
    "canUseTools": true,
    "canTriggerApprovals": true,
    "canCallOtherAgents": false,
    "canExecuteCode": true
  },
  "schedule": {
    "type": "cron",
    "expression": "0 9 * * 1",  // Every Monday at 9am
    "timezone": "Pacific/Auckland"
  }
}

Use cases: Report generation, data sync, maintenance tasks, notifications

Creating an Agent: Step by Step

Navigate to Agent Creation

Go to Dashboard → Agents → Create New Agent

Choose a Template or Start Fresh

Templates provide pre-configured agents for common use cases. See Agent Templates for available options.

Configure Basic Settings

NameUnique identifier (e.g., content-writer-v1)

Display NameHuman-readable name for UI

DescriptionWhat this agent does

Agent Typeassistant, coordinator, specialist, scheduled

Select AI Model

Choose the LLM that powers your agent:

Claude 3.5 Sonnet

Best for complex reasoning, writing

GPT-4o

Fast, capable, multimodal

Claude 3 Haiku

Fast, cost-effective for simple tasks

GPT-4o Mini

Budget-friendly, good performance

Write System Prompt

The system prompt defines your agent's personality, expertise, and constraints. This is the most important configuration.

Assign Tools & Capabilities

Select which tools your agent can use. See Tools & Integrations for available options.

Test & Deploy

Use the chat interface to test, then activate your agent to make it available for workflows.

Best Practices

Do: Single Responsibility

Create focused agents that do one thing well. A “Content Writer” agent should only write content, not also handle scheduling and analytics.

Do: Explicit Constraints

Define clear boundaries in your system prompt. Tell the agent what it should NOT do as clearly as what it should do.

Don't: Over-provision Tools

Give agents only the tools they need. An agent with access to all tools is harder to control and more likely to make mistakes.

Don't: Skip Testing

Always test agents thoroughly before deploying to production. Use edge cases and adversarial inputs to find weaknesses.

Continue Learning

Agent Configuration

Deep dive into model settings, prompts, and parameters

Agent Validation

Learn how agents are validated before deployment

Agent Templates

Pre-built agent configurations for common use cases

Agent Guardrails

Safety constraints and approval requirements