In this challenge, you will create a working agent connected to your own knowledge graph, test it, and save it.
Use any AuraDB instance with data loaded. The steps are the same regardless of your schema.
Before you start
Check that both features are enabled in your organization settings:
-
Generative AI assistance is on
-
Aura Agent is on
You also need to be a Project Admin to create an agent.
Step 1: Choose your dataset
Decide which AuraDB instance to use. It must already have nodes and relationships loaded.
Don’t have data yet?
Load the Northwind sample dataset — a food distribution graph with customers, orders, products, and suppliers. The example prompts in this lesson use Northwind.
You can learn more about creating an instance in Aura Fundamentals.
Step 2: Create the agent
Go to Data Services → Agents.
If this is your first time creating an agent, you will see options to Create with AI or Create from scratch. If you have already created an agent, click Create Agent and select Create with AI.
In the dialog:
-
Select the AuraDB instance you chose in Step 1.
-
Write a prompt that describes what the agent should do. Define its role, what data it can answer questions about, and that it should decline off-topic or harmful requests.
Here is an example prompt for a customer service agent for Northwind Traders:
You're an expert customer service agent for Northwind Traders, a food distribution company.
Your role is to answer questions about customers, orders, products, categories, and suppliers.
Decline off-topic or harmful requests.After entering the prompt, click Create. The agent generates tools based on your instance schema and the prompt you provided.
Step 3: Preview and test the agent
The agent configuration page will open with an AI generated description and a set of tools based on the prompt you provided.
Review the Name and Description fields, and modify if necessary.
Use the preview panel to ask questions that your graph can answer. Try at least two or three to exercise different tools.
If you are using Northwind, try:
-
Which are the top 5 most ordered products?
-
Who has ordered Pavlova repeatedly?
-
What products do they order most?
Trace the reasoning
Every agent response has a Thought section. Expand it to see the execution trace:
-
Reasoning: the LLM’s internal thought before it acts. Read this to understand why a specific tool was chosen.
-
Applying agent tool: what was sent to the tool and what came back — the tool name, parameter values, and raw output (Cypher query and results, or vector similarity results).
Compare the reasoning text between responses to see how the LLM explains its tool selection depending on the question.
Verifying the generated Cypher
If the Text2Cypher tool produces Cypher with incorrect relationship types or node labels, try adding the correct nodes and relationship types to the tool description.
Step 4: Save the agent
When you are satisfied with the agent’s behavior, click Create Agent to save it.
You now have a working agent that you can use to answer questions about your data using natural language.
Summary
In this challenge, you created a working agent connected to your knowledge graph, tested it, and traced the reasoning.
In the next module, you will design and build an agent with Cypher Template and Text2Cypher tools.