<code>pgvector</code> is a PostgreSQL extension that provides powerful functionalities for working with high-dimensional vectors. It introduces a dedicated data type, operators, and functions that enable efficient storage, manipulation, and analysis of vector data directly within the PostgreSQL database. If you’re looking for a vector database, know that PostgreSQL is all you need.
The <a href="https://console.cloud.timescale.com/signup/?utm_source=devto&amp;utm_medium=referral&amp;utm_campaign=dev-2025&amp;utm_content=timescale-sign-up-page/" target="_blank" rel="noopener nofollow">Timescale Cloud platform</a> includes two new open-source PostgreSQL extensions developed by the Timescale Team that complement pgvector: <a href="https://github.com/timescale/pgai" target="_blank" rel="noopener nofollow">pgai</a>, which brings more AI workflows to PostgreSQL; and pgvectorscale, which enables developers to build more scalable AI applications, with higher performance embedding search and cost-efficient storage. You can <a href="https://www.timescale.com/blog/making-postgresql-a-better-ai-database?utm_source=devto&amp;utm_medium=referral&amp;utm_campaign=dev-2025&amp;utm_content=making-postgresql-a-better-ai-database/" target="_blank" rel="noopener nofollow">learn more about these two extensions in this article</a>.
Let’s walk through the main features and use cases. We’ll also cover a few examples of analyzing relational and time-series data in TimescaleDB.
<h2>Understanding pgvector: Enabling Vector Operations</h2>
The pgvector extension is a powerful tool for storing, modifying, and querying vectors. This functionality enables applications such as similarity and semantic search, retrieval augmented generation, image search, recommendation systems, natural language processing (NLP), and computer vision. To kick off our pgvector exploration, let’s take a look at vectors themselves.
<h2>What are vectors?</h2>
Vectors refer to mathematical representations of data points in multidimensional space. They are typically arrays or lists of numerical values that capture an entity’s essential features or attributes. In machine learning and data science, vectors encode information in a way that enables efficient computation and analysis.
Vectors can represent words, sentences, or documents, allowing for semantic search where similar meanings are identified even if the exact words differ. This can be used to find documents with similar content or answer queries based on the context of the text. By converting visual data into vectors, vector databases (such as PostgreSQL with the pgvector extension) enable searching for similar images or videos. This can be useful in applications like facial recognition, object detection, and content-based image retrieval.
In time series and other types of data, vectors can represent normal behavior patterns, allowing the detection of anomalies by identifying vectors that significantly deviate from the norm.
Here are some of the key features and use cases of pgvector:
<ul>
<li>
Vector storage: The pgvector extension lets you store high-dimensional vectors directly in PostgreSQL tables. It provides a dedicated data type for vector representation, allowing efficient storage and retrieval of vector data.
</li>
<li>
Similarity search: With pgvector, you can perform <a href="https://www.timescale.com/learn/vector-search-vs-semantic-search?utm_source=devto&amp;utm_medium=referral&amp;utm_campaign=dev-2025&amp;utm_content=vector-search-vs-semantic-search/" target="_blank" rel="noopener nofollow">similarity searches</a> based on vector similarity metrics such as cosine similarity or Euclidean distance. Searching for similar vectors facilitates applications like content-based recommendation systems, k nearest-neighbor search, and clustering.
</li>
<li>
Semantic search: Large language model (LLM) embeddings are a way to create vectors from other types of data (e.g., text, images, etc). These embeddings represent the meaning of the underlying data. Therefore, using similarity search to find similar vectors returns data with similar semantic meaning or content, effectively identifying data points that share common themes or topics regardless of their original format (text, images, etc).
</li>
<li>
Natural Language Processing (NLP) and text analysis: Vector embeddings are powerful as they capture the underlying meaning of text. Techniques like word or document embeddings enable you to capture the ideas expressed in the text. You can then perform vector-based operations like similarity search, clustering, or classification on textual data.
</li>
<li>
Computer vision: The pgvector extension can handle vector representations of images and enable similarity-based image search. You can convert images to vector representations using convolutional neural networks (CNN) or image embeddings. These capabilities allow you to perform content-based image retrieval, image similarity matching, object identification, and image clustering within the database.
</li>
</ul>
<img src="https://dev-to-uploads.s3.amazonaws.com/uploads/articles/7oon6oww7k0c8kfop2yj.jpg" alt="pgvector makes postgres a vector database">
Pgvector stores vectors that represent semantic meaning. Thus things that are meaningfully similar are “closer” together in the vector space. This allows people searching for SUVs to find cars and trucks instead of watermelons and apples, even though SUVs and trucks are different words.
<h2>Using Pgvector for Vector Data</h2>
By utilizing the pgvector extension, developers can efficiently store and query vector data within PostgreSQL, making it easier to build generative AI applications with LLMs, as well as AI applications that require similarity search, recommendation systems, NLP, and other tasks that involve working with vectors.
<h3>Installation</h3>
You can use pgvector on a cloud database service in Timescale or a self-hosted PostgreSQL instance:
<ul>
<li>
Timescale’s cloud database instances come with <code>pgvector</code> pre-installed. So, all you have to do is run <code>CREATE EXTENSION</code> or connect to the PostgreSQL service with <a href="https://github.com/timescale/python-vector" target="_blank" rel="noopener nofollow">Timescale’s Python client library</a>. <a href="https://console.cloud.timescale.com/signup/?utm_source=devto&amp;utm_medium=referral&amp;utm_campaign=dev-2025&amp;utm_content=timescale-sign-up-page/" target="_blank" rel="noopener nofollow">Get started with <code>pgvector</code> on Timescale</a>.
</li>
<li>
For self-hosting, you need to install the pgvector extension on your server. You can follow the pgvector installation instructions <a href="https://github.com/pgvector/pgvector?tab=readme-ov-file#installation/" target="_blank" rel="noopener nofollow">for Linux, MacOS, and Windows</a>. Once installed, you can enable the extension in your database using the <code>CREATE EXTENSION</code> command: <code>CREATE EXTENSION vector;</code>
</li>
</ul>
<h3>Vector data type</h3>
pgvector introduces a new data type called <code>vector</code> representing a high-dimensional vector. You can define <code>vector-type</code> columns in your database tables to store vector data. For example, you can have a table called <code>documents</code> with an <code>embedding</code> column of type <code>vector</code> to store LLM embeddings as vectors for each document.
<pre><code class="language-SQL">CREATE TABLE documents (
 id BIGINT PRIMARY KEY GENERATED ALWAYS AS IDENTITY,
 content TEXT,
 author_id BIGINT,
 embedding VECTOR(1538)
);
</code></pre>
<h3>Indexing and vector search</h3>
pgvector provides indexing mechanisms optimized for approximate nearest neighbor search over vector data. You can create an index on a vector column using the <code>CREATE INDEX</code> command, specifying the <code>hnsw</code> index type and a <code>vector_cosine_ops</code> operator class. This enables fast similarity searches on vector data using the cosine distance similarity metric. You can then use the <code>&lt;=&gt;</code> operator to perform similarity searches in your queries:
<pre><code class="language-SQL">CREATE INDEX ON documents USING hnsw (embedding vector_cosine_ops);

--find the closest 5 elements to a given query
SELECT * FROM documents ORDER BY embedding &lt;=&gt; &#x27;[10.5, 11.0,...]&#x27; LIMIT 5;.
</code></pre>
<h3>Vector functions</h3>
pgvector has a set of built-in functions to manipulate and perform operations on vector data. These functions allow you to calculate vector similarities, perform vector arithmetic, and more. For example, you can use the <code>cosine_distance</code> function to calculate the cosine similarity between two vectors. Other useful functions are <code>vector_norm()</code> to get the Euclidean norm and <code>vector_dims()</code> to determine how many dimensions a vector contains.
<h3>Vector aggregates</h3>
pgvector has the <code>avg(vector)</code> and <code>sum(vector)</code>
aggregate functions for calculating analytics on vectors.
<h3>Integration with other PostgreSQL features</h3>
Pgvector seamlessly integrates with other features of PostgreSQL, such as transaction management, query optimization, and security. This integration enables you to leverage the power of PostgreSQL for complex data processing tasks. Joining pgvector data with other data types, like relational, time-series, and geospatial data, is a powerful way to enrich your vector queries’ data. As a simple example, you could return author information for documents found via similarity search:
<pre><code class="language-SQL">WITH matching_docs as (
 --find 5 closest matches 
 SELECT * 
 FROM documents 
 ORDER BY embedding &lt;=&gt; &#x27;[10.5, 11.0,...]&#x27; 
 LIMIT 5
)
SELECT d.content, a.first_name, a.last_name
FROM matching_docs d 
INNER JOIN author a ON (a.id = d.author_id).
</code></pre>
<h2>Why Use Pgvector With Timescale?</h2>
Timescale is PostgreSQL++ engineered for time-series data, events, and analytics. It provides a robust foundation for storing, retrieving, and <a href="https://www.timescale.com/blog/time-series-analysis-what-is-it-how-to-use-it?utm_source=devto&amp;utm_medium=referral&amp;utm_campaign=dev-2025&amp;utm_content=time-series-analysis/" target="_blank" rel="noopener nofollow">analyzing large volumes of time-series data</a> thanks to its time-based aggregations, <a href="https://docs.timescale.com/use-timescale/latest/data-retention/create-a-retention-policy/" target="_blank" rel="noopener nofollow">data retention policies</a>, and features like <a href="https://docs.timescale.com/use-timescale/latest/hypertables/about-hypertables/" target="_blank" rel="noopener nofollow">hypertables</a> and <a href="https://www.timescale.com/learn/postgresql-materialized-views-and-where-to-find-them?utm_source=devto&amp;utm_medium=referral&amp;utm_campaign=dev-2025&amp;utm_content=postgresql-materialized-views/" target="_blank" rel="noopener nofollow">continuous aggregates</a>.
Many real-world AI applications have retrieval or analytical requirements that include

PostgreSQL insights for every developer journey. Get tips and tutorials on database performance, query optimization, time-series data, and more—from real-time analytics and IoT data management to events and vector embeddings. 


TigerData (Creators of TimescaleDB)

The pgvector extension for PostgreSQL enables efficient storage, manipulation, and analysis of high-dimensional vector data. It supports applications like similarity search, recommendation systems, NLP, and computer vision. Timescale Cloud’s new extensions, pgai and pgvectorscale, further enhance PostgreSQL's capabilities for AI workflows and scalable AI applications. The tutorial covers installation, vector data types, indexing, vector functions, and integration with PostgreSQL features.

[TUTORIAL] How to Turn Postgres Into A Vector Database

Continue to the full <a href="%5Burl%5D(https://www.timescale.com/learn/postgresql-extensions-pgvector)" target="_blank" rel="noopener nofollow">Postgres as a VectorDB</a> tutorial