SME Agent Setup

This guide walks you through configuring and customizing AI agents for your business. You'll learn how to choose the right agent types, configure their capabilities, and set appropriate autonomy levels.

Choosing Your First Agent

FAOS offers several pre-built agent templates optimized for SME use cases:

Agent Type	Best For	Key Capabilities
Customer Service	Support inquiries	FAQ answers, ticket routing, status updates
Operations	Daily tasks	Scheduling, reminders, data entry
Research	Information gathering	Web research, summarization, reports
Sales Assistant	Lead management	Follow-ups, proposals, CRM updates

Recommended Starting Point

For most SMEs, we recommend starting with a Customer Service Agent:

• Immediate ROI through reduced response times
• Low risk—can be supervised initially
• Builds trust in the system before expanding

Agent Configuration

Step 1: Basic Information

Navigate to Agents → + New Agent and configure:

# Basic Agent Configuration
name: "Support Assistant"
description: "Handles customer inquiries and support requests"
template: "customer-service"
industry: "your-industry"  # Loads relevant ontology

Step 2: Set Autonomy Level

Choose how independently your agent operates:

Level	Name	What It Means	When to Use
L1	Assisted	Agent suggests, you execute	New agents, sensitive tasks
L2	Supervised	Agent executes, you approve first	Default recommendation
L3	Autonomous	Agent executes within guardrails	Proven, routine tasks
L4	Proactive	Agent anticipates and acts	Advanced, high trust

Start with L2

We recommend starting with L2 (Supervised) for all new agents. This lets you review actions before they execute while the agent learns your preferences.

Step 3: Configure Guardrails

Guardrails define what your agent can and cannot do:

// Example guardrails configuration
const guardrails = {
  // Actions requiring your approval
  requireApproval: [
    "refund_above_50",
    "discount_above_10_percent",
    "account_modifications",
    "external_communications"
  ],

  // Actions never allowed
  blocked: [
    "share_customer_pii",
    "make_promises_not_in_policy",
    "access_financial_data"
  ],

  // Rate limits
  limits: {
    actionsPerHour: 50,
    emailsPerDay: 100,
    apiCallsPerMinute: 10
  }
};

Step 4: Knowledge Base

Give your agent the information it needs:

1. FAQ Documents: Upload your FAQ or knowledge base
2. Product Information: Catalogs, pricing, specifications
3. Policies: Return policy, terms of service, guidelines
4. Templates: Response templates, email formats

📁 Knowledge Base Structure
├── 📄 faq.md
├── 📄 return-policy.md
├── 📄 pricing-guide.pdf
├── 📄 response-templates.md
└── 📄 escalation-procedures.md

Agent Customization

Personality & Tone

Configure how your agent communicates:

personality:
  tone: "friendly-professional"  # Options: formal, casual, friendly-professional
  greeting: "Hi! I'm here to help with any questions."
  signoff: "Is there anything else I can help you with?"
  empathy_level: "high"  # Acknowledge customer frustration

communication:
  response_length: "concise"  # concise, detailed, adaptive
  use_emojis: false
  language: "en-US"

Business Hours

Set when your agent is available:

availability:
  timezone: "America/New_York"
  business_hours:
    monday: "9:00-17:00"
    tuesday: "9:00-17:00"
    wednesday: "9:00-17:00"
    thursday: "9:00-17:00"
    friday: "9:00-17:00"
    saturday: "closed"
    sunday: "closed"

  after_hours:
    action: "collect_info"  # Options: offline, collect_info, limited_service
    message: "Thanks for reaching out! We'll respond first thing tomorrow."

Escalation Rules

Define when to involve a human:

escalation:
  triggers:
    - sentiment: "angry"
    - keywords: ["lawsuit", "lawyer", "attorney", "sue"]
    - repeat_contacts: 3  # Same issue, 3+ contacts
    - vip_customers: true

  routing:
    urgent: "support-manager@company.com"
    billing: "billing@company.com"
    technical: "tech-support@company.com"
    default: "support@company.com"

Testing Your Agent

Before going live, test thoroughly:

1. Sandbox Testing

Use the built-in sandbox to simulate conversations:

You → "I want to return my order"
Agent → "I'd be happy to help with your return! Could you please provide your order number?"
You → "Order #12345"
Agent → [Looks up order] "I found your order from last week. Our return policy allows returns within 30 days. Would you like me to initiate a return?"

2. Test Scenarios

Run through common scenarios:

• Simple FAQ question
• Order status inquiry
• Return/refund request
• Complaint handling
• Out-of-scope question (should escalate)
• After-hours inquiry

3. Edge Cases

Test boundary conditions:

• Angry customer language
• Multiple questions in one message
• Request requiring approval
• Request that should be blocked

Monitoring & Optimization

Dashboard Metrics

Track your agent's performance:

Metric	Target	Description
Response Time	< 30 seconds	Time to first response
Resolution Rate	> 70%	Issues resolved without human
Customer Satisfaction	> 4.0/5	Post-interaction rating
Escalation Rate	< 20%	Conversations requiring human

Continuous Improvement

1. Review Escalations: Analyze why conversations were escalated
2. Update Knowledge: Add new FAQs based on common questions
3. Adjust Guardrails: Relax or tighten based on performance
4. Increase Autonomy: Graduate from L2 to L3 for proven tasks

Common Configuration Examples

E-commerce Support Agent

const ecommerceAgent = {
  name: "Shop Assistant",
  template: "customer-service",
  industry: "retail",

  capabilities: {
    orderLookup: true,
    returnProcessing: true,
    productRecommendations: true,
    inventoryCheck: true
  },

  guardrails: {
    autoRefundLimit: 50,
    requireApprovalFor: ["refund_above_50", "price_match"],
    blockedActions: ["change_shipping_address"]
  }
};

Professional Services Agent

const servicesAgent = {
  name: "Client Coordinator",
  template: "operations",
  industry: "professional-services",

  capabilities: {
    scheduling: true,
    documentCollection: true,
    appointmentReminders: true,
    intakeProcessing: true
  },

  guardrails: {
    requireApprovalFor: ["schedule_consultation", "send_contract"],
    blockedActions: ["provide_advice", "quote_pricing"]
  }
};

Next Steps

• SME Workflows - Create automated workflows
• Industry Ontologies - Explore domain knowledge
• API Reference - Technical integration

---

Need Help?

Join our Discord community for real-time support from other SME users and the FAOS team.