Guide for Agentic AI Agents

What is an Agentic AI Agent?

An Agentic AI Agent is advanced conversational intelligence powered by a Large Language Model (LLM), designed to operate autonomously while adapting dynamically in real time to context, persona, and instructions. Unlike traditional intent- or rule-based systems, these AI Agents go beyond scripted responses — demonstrating flexibility, problem-solving capabilities, and proactive decision-making.

In this guide, you will create a Logistics AI Agent to answer general questions and assist with parcel inquiries.

Prerequisites

Before creating an AI Agent, configure the following:

• LLM for language generation, for example, OpenAI GPT-4o.
• LLM for embeddings generation, for example, OpenAI ada-002.

Create your AI Agent's Persona

Set up the essential characteristics that define your AI Agent's behavior and role.

1. Go to Build > AI Agents.
2. Define general settings:
   • Name — for example, Peter Parcel.
   • Description — for example, A logistics expert passionate about efficiency.
   • Avatar.
3. Set up speaking style (tone, verbosity, formality) and voice.
4. (Optional) Assign knowledge and add instructions.
5. Ensure security settings are configured properly.
6. Assign a job and click Create to test.

AI Agent Creation Wizard

Assign a Job to your AI Agent

Specify the role and tasks your AI Agent will perform within a Flow.

1. Go to Build > Flows and open the Flow assigned to the AI Agent.
2. Ensure an AI Agent Node is present and assign the AI Agent Persona to the Node.
3. Define job details:
   • Job Title — for example, Logistics Support Specialist.
   • Description and Instructions.
4. Experiment and refine settings in the Interaction Panel.

AI Agent Job Configuration

Enhance your AI Agent with Knowledge

Generative models may lack company-specific knowledge. To improve accuracy, use Cognigy Knowledge AI (RAG) by integrating external knowledge sources.

1. Ensure language and embedding models are set up. See Prerequisites.
2. Go to Knowledge AI in the left-side menu.
3. Upload relevant documents, for example, PDF files.
4. Assign knowledge to the AI Agent in its configuration.
5. Customize knowledge settings per AI Agent Node.

Improve AI Agent Skills with Tool Actions

Enable your AI Agent to perform specific tasks and integrate with external systems.

Example Tool Actions

• track parcel — retrieve tracking details.
• escalate to a human agent — route requests to a human agent.

Create a Tool Action

1. Click Add Tool Action next to your AI Agent Node.
2. Set Tool ID, for example, track_parcel, and description.
3. Enable parameters, for example, add tracking_code as a string.
4. Use Resolve Tool Action to finalize the AI Agent's response.

AI Agent Tool Action

Enable AI Agent Memory

Configure memory settings to personalize interactions and retain user details.

Long-Term Memory

• Stores user-specific details across conversations, for example, preferred name, interests.
• Managed via the Contact Profile.

Short-Term Memory

• Stores session-specific data.

• Erased after a conversation ends.

• Configure the AI Agent Node settings in the Memory Handling section.

• Set Long-Term Memory Injection to Inject full Contact Profile.
• (Optional) Add an Update Profile Node to simulate stored user data.

AI Agent Memory Configuration

Enable Image Recognition

AI Agents can process images if a vision-capable LLM is used.

1. Activate Process Images in the AI Agent Node.
2. Enable Attachment Upload in the Webchat v3 Endpoint settings.
3. Use a storage provider, for example, AWS or Azure.

Deploy your AI Agent

Cognigy.AI Endpoints allow AI Agents to interact with external users via chat and voice channels.

Deploy an AI Agent via Webchat:

1. Go to Deploy > Endpoints.
2. Add a new Endpoint and link it to the Flow.
3. Use Open Demo Webchat to test the AI Agent.

What's Next?

Explore advanced features, test different configurations, and refine responses based on real-world interactions.