Guideline-Driven Agents

1. Creating an Agent

To create a new agent in the Flowk Dashboard:

1. Navigate to AI Studio in the sidebar.
2. Click Create Agent and give your agent a name (e.g., "Support Agent").
3. Describe the agent's role, purpose, tone and any other instructions which defines its behavior.

2. Writing Guidelines

Guidelines are the core of your agent's behavior. Each guideline consists of a condition and an action:

guideline "greet the customer" {
  when a new conversation starts
  then respond with a friendly greeting and ask how you can help
}

guideline "verify order before refund" {
  when the customer requests a refund
  then ask for their order number before proceeding
}

guideline "escalate to human" {
  when the customer asks to speak to a human
    or the customer uses profanity
    or the customer mentions a legal issue
  then escalate the conversation to a human agent
}

Guidelines are evaluated based on specificity — more specific rules override broader ones. To keep things organized, you can group related guidelines into Journeys for different contexts such as “Sales”, “Support”, or “Billing”.

Guideline Best Practices

• Begin with broad behavioral guidelines such as tone, greetings, and escalation handling.
• Introduce business-specific rules incrementally and test each one as you go.
• Use positive instructions — clearly state what the agent should do rather than what it should avoid.
• Keep each guideline focused on a single condition and its corresponding action.
• Validate your guidelines using the built-in conversation simulator.
• Define your agent's guidelines and glossary terms, and connect tools where needed (e.g., human handover webhooks, RAG systems, or external APIs).
• Use the AI Wizard/Launchpad to automatically generate guidelines and glossary terms from your documents or URLs.
• Save your agent and publish it when ready.
• Continuously monitor performance through the hybrid inbox and refine guidelines based on real conversation insights.

3. Journeys

Journeys help you organize guidelines rules into logical groups. They are automatically triggered based on the conversation context. For example:

• General Behavior — Covers greetings, tone, and escalation rules.
• Product Support — Includes troubleshooting steps and return policies.
• Billing — Handles payment issues, invoice requests, and subscription changes.

4. Testing Agents

Before deploying your agent to a live channel, use the built-in conversation chat widget to test its behavior. Enter sample customer messages and observe how the agent responds based on your guidelines. The chat widget helps to validate the agent's response, making it easy to debug unexpected behavior.

5. Agent Monitoring

Once your agent is live, you can monitor its performance from the dashboard. Key metrics include total conversations handled, escalation rate, average response time, and guideline adherence. Use this data to continuously refine your guidelines and improve your agent's effectiveness.