The Similarity Search tool finds nodes in your graph using semantic similarity rather than exact property matches.
In this lesson, you will learn:
-
When to use the Similarity Search tool
-
What your database needs before you can use it
-
How to configure a Similarity Search tool in the Aura Console
When to use Similarity Search
A Similarity Search tool finds nodes whose vector embeddings are closest to the embedding of the user’s question.
Use it when the user is looking for entities that are related or similar rather than a specific known value:
-
"What products are similar to Chai?"
-
"Find documents about data governance."
-
"Which terms are closest in meaning to this description?"
Similarity Search is a natural complement to Cypher Templates. The Similarity Search tool can find semantically related nodes, and a Cypher Template can then retrieve structured data connected to those nodes.
Prerequisites
Before adding a Similarity Search tool, your AuraDB instance must have:
-
Vector embeddings stored on the nodes you want to search — generated by one of the supported embedding providers (see below).
-
A vector index built on those embeddings.
Supported embedding providers
Aura Agent supports the following text embedding providers and models:
| Provider | Models |
|---|---|
OpenAI (via Azure OpenAI) |
|
Vertex AI |
|
The agent embeds the user’s question using the same model used to generate the stored embeddings, ensuring the vector spaces align.
For more information, see the Cypher Manual — Vector indexes and the Aura Agent documentation.
Using your own data
The Northwind sample dataset used in this course does not include vector embeddings. To use a Similarity Search tool, you need a database with embeddings already loaded.
Configuring a Similarity Search tool
In the Aura Console, add a Similarity Search tool from the agent configuration: Add Tool → Similarity Search.
The tool has two configuration fields:
| Field | Description |
|---|---|
Vector index name |
The exact name of the vector index in your AuraDB instance. |
Top K |
The number of similar nodes to return. Start with 5–10; increase if you need more candidates for the LLM to reason over. |
Give the tool a clear Name and a Description that tells the agent when to use it. The description should describe the value returned by the tool - for example: "Find products semantically similar to the user’s description."
Summary
The Similarity Search tool finds nodes by semantic similarity using vector embeddings. It requires a vector index and embeddings generated by a supported provider in your AuraDB instance. Configure it with the index name and a Top K value.
In the next module, you will publish your agent and connect it through MCP.