GraphAcademy

Neo4j & Generative AI Hands-On Workshop

Generative AI

Retrievers

Agents

Aura Agents

No-code Agents

The approach used so far in this module has involved writing code to create agents using Python and LangChain.

Neo4j Aura also provides a no-code interface to create agents that demonstrate GraphRAG through Aura Agents.

What are Aura Agents?

The key features of Aura Agents are:

  • No-code creation - Build agents through a web interface

  • GraphRAG integration - Leverage your knowledge graph structure

  • Multiple tool types - Combine different query approaches

  • API accessibility - Make agents available via REST endpoints

Creating an Agent

To create an agent, select Agents from the left hand menu and click Create. You can configure your agent with a:

  • Name - A clear, descriptive name for the agent

  • Description - A brief explanation of the agent’s purpose

  • Prompt Instructions - Instructions passed to the LLM to provide context for the agent and how it should behave

  • Target instance - The Neo4j Aura instance that the agent will connect to

  • Visibility - The agent can be made available internally for members of the Aura project, or available externally via an API endpoint.

Setting up our EDGAR SEC Filings agent:

  • Name: EDGAR SEC Filings Agent

  • Description:

An agent that can answer questions about the EDGAR SEC filings.

  • Prompt Instructions:

You are an expert in the EDGAR SEC filings. You have access to a graph database containing information about companies, executives, financial metrics, and business risks extracted from the EDGAR SEC filings.

  • Target instance: Select Your instance

  • Visibility: Internal

Agent Tool Types

Aura Agents support three different tool types:

  • Similarity Search Tools - Vector-based semantic search

  • Cypher Template Tools - Predefined queries with parameters

  • Text-to-Cypher Tools - Natural language to Cypher translation

These tools are used to provide an LLM with the context required to perform the task at hand.

Similarity Search Tools

Purpose: Find semantically similar content using vector embeddings.

Best for:

  • Document search

  • Content discovery

  • Finding similar clauses or terms

  • Semantic matching

Example Query: "What are the risks that Apple faces?"

Configuration:

  • Name: Risk Finder

  • Description: Find companies that face a type of risk.

  • Embedding Provider: OpenAI

  • Embedding Model: text-embedding-ada-002

  • Index: chunkEmbeddings

  • Top K: 10

Cypher Template Tools

Purpose: Execute predefined Cypher queries with user-provided parameters.

Best for:

  • Common, repeated questions

  • Deterministic results with consistent performance

  • Complex queries using full Cypher feature set

  • Well-defined business logic patterns

Connection to Code: These tools implement the same pattern as direct Cypher queries you’ve written, but packaged for reuse by the agent.

Example Query: "What companies are owned by BlackRock Inc.?"

Configuration:

  • Name: Get holdings for Asset Manager

  • Description: Find all companies owned by a asset manager by their name.

  • Parameters:

    • asset_manager, a string - The full name of the asset manager.

  • Cypher Template:

MATCH (owner:AssetManager {managerName: $asset_manager})
RETURN owner.managerName AS managerName,
    [ (owner)-[:OWNS]->(company) | company.name] AS owned_companies

Text-to-Cypher Tools

Purpose: Convert natural language questions into Cypher queries dynamically.

Best for:

  • Catch-all for unforseen questions

  • Well-defined questions that map directly to the schema

  • Ad-hoc analysis

  • Questions you haven’t created templates for

Example Queries:

"Which documents mention the metric 'net loss'?"
"List the asset managers in ascending order of the number of companies they own shares in."

Configuration:

  • Name: Catch-all data tool

  • Description:

    A tool that can answer any question about the graph that cannot be specifically answered by the other tools.

Testing the agent

Clicking an agent in the left hand pane will reveal a chat interface that you can use to test the agent.

You can test the tools in sequence.

  1. List the top 5 asset managers by name in ascending order. - Text-to-Cypher tool

  2. Which companies does "ALLIANCEBERNSTEIN L.P." hold shares in? - Cypher Template tool

  3. What are the risks that Apple faces? - Similarity Search tool

The agent will choose the best tool for each question.

The UI will display the time taken to generate each response.

Expand the Thought for…​ section to view the steps and tool calls taken to generate the answer.

Example Agents

You can find example agents in the Neo4j Aura Agents GitHub repository on GitHub.

Continue

Why not experiment yourself by creating your own Aura Agent using your own Neo4j Aura instance?

You can learn more about getting started with Neo4j Aura in our Aura Fundamentals course.

Summary

In this lesson, you learned about Neo4j Aura Agents and how to create no-code agents for your knowledge graph:

Key Concepts:

  • Aura Agents are a no-code interface for creating GraphRAG-powered chatbots

Tool Types:

  • Cypher Template Tools - Execute predefined queries for known patterns and specific lookups

  • Text-to-Cypher Tools - Convert natural language to Cypher for flexible exploration

  • Similarity Search Tools - Use vector embeddings for semantic content discovery

Example Agents: * Legal - Contract Review Agent * Financial Services - Know Your Customer Agent

Chatbot

How can I help you today?