Vector Search Database

Controlled node

Overview

The Vector Search Database node performs semantic similarity searches on vector embeddings stored in a project database. It compares an input query vector against vectors stored in a specified column using various distance metrics (cosine similarity, inner product, L2 distance, or L1 distance) and returns the most similar rows.

This node is essential for implementing semantic search, recommendation systems, or any workflow requiring similarity matching on high-dimensional vector data (such as text embeddings or image embeddings).

Distance Metrics

The node supports multiple distance metrics for calculating similarity:

Metric	Description	Best For
Cosine Similarity	Measures the cosine of the angle between two vectors	Text embeddings, semantic similarity where magnitude doesn't matter
Inner Product	Standard dot product of vectors	When vectors are normalized or for specific ML models
L2 Distance	Euclidean distance between vectors	Physical distance measurements, image embeddings
L1 Distance	Manhattan distance between vectors	Robust to outliers, sparse vectors

Vector Column Setup

Ensure your database table has a column containing vector data (typically generated using the Get Embeddings node or similar). The default column name is embedding, but this can be configured.

Inputs

Input	Type	Description	Default
Run	Event	Triggers the vector search operation	-
Database ID	Text	The ID of the project database to search	-
Table ID	Text	The ID of the table within the database	-
Column ID	Text	The name of the column containing vector embeddings	embedding
Vector	Data	The query vector (array of numbers) to search for	-
Limit	Number	Maximum number of similar results to return	10
Distance Metric	Enum	The similarity metric to use: cosine, innerProduct, l2, or l1	cosine

Outputs

Output	Type	Description
Done	Event	Fires when the search operation completes successfully
Result	Data	Array of matching rows, typically including the row data and similarity scores

Runtime Behavior

When triggered by the Run event, the node executes a vector similarity search against the specified database table. It validates that:

• The database and table exist and are accessible
• The vector column exists in the table
• The input vector is a valid non-empty array of numbers
• The limit is a positive number

The node returns an array of rows sorted by similarity (closest matches first). Each result typically includes the original row data plus a score indicating the distance/similarity to the query vector.

If the input vector is invalid, empty, or the specified column doesn't exist, the node returns an error object in the Result output.

Example Usage

Scenario: Finding similar documents based on a text query

1. Use AI Write or Get Embeddings to generate an embedding vector for your search query
2. Connect the vector output to the Vector input of Vector Search Database
3. Configure the node with your database ID, table ID, and the column name where document embeddings are stored (e.g., embedding)
4. Set the Limit to control how many similar documents to retrieve (e.g., 5)
5. Choose Cosine Similarity for text embeddings
6. When triggered, the node returns the most semantically similar documents from your database

Typical Workflow:

[Get Embeddings] → [Vector Search Database] → [For Each]
       ↑                    ↓                      ↓
   Query Text         Similar Documents      Process Results

Performance Considerations

Vector searches on large datasets may take time depending on the database size and whether vector indexes are configured. For production use with large tables, ensure your database has appropriate vector indexing enabled.