Enable javascript in your browser for better experience. Need to know to enable it? Go here.

Vector databases

Vector databases store data in a way that captures the multidimensional nature of the data they contain. While ‘traditional’ databases store data in structured rows and columns, the ‘vectors’ of vector databases are essentially lists of numbers which act as representations of a given data point. 


These vectors are often used to capture complex information, like the meaning of words or images, in a format that computers can easily understand and compare. This type of database helps in quickly finding similar items, making it useful for tasks like recommendations or image searches.

What is it?

Vector databases are a unique kind of database in which data is stored as rich mathematical codes.

What’s in it for you?

Vector databases can better represent the multi-dimensional nature of unstructured data, which can support use cases including search and AI applications.

What are the trade-offs?

Vector databases require a certain level of expertise and aren’t appropriate for more basic use cases.

How is it being used?

Vector databases are a critical part of the new wave of AI-backed applications.

What are vector databases?


Vector databases are a type of database that stores data in a way that captures the richness of the relationships between data points. It does this by storing data as ‘vectors’. These are representations of the data across many different dimensions.


This design makes it a particularly useful database for things like search and AI. It means unstructured data can be indexed and discovered incredibly quickly compared to other kinds of databases.

What’s in it for you?


There are a number of advantages of vector databases. For instance, they can help you...


  • Enhance customer experience and improve search. Because of the way vector databases store and index data, vector databases can support better recommendation engines and improve the speed and quality of search.

  • Supercharge AI initiatives. Many AI applications rely on understanding the relationships between data points. Vector databases excel at this, unlocking new possibilities for AI-powered tools in areas as diverse as product development, fraud detection and personalization.


Ultimately, vector databases help you unlock value from unstructured data such as text, images, videos. While traditional databases struggle to capture the richness of such data, vector databases make it easier to organize and extract meaning from diverse formats. This means they can be leveraged for analysis.

What are the trade-offs of vector databases?


Vector databases have a number of drawbacks:


  • They're more complex compared to other kinds of database. This means there's a steeper learning curve. Your team might need additional training.

  • They excel at finding similar things, but struggle with queries that involve specific and exact criteria, like, say “blue jeans, size small.” For tasks that require complex filtering or sorting, traditional databases might be a better fit.

  • Vector databases perform well with high-dimensional data (like images and text embeddings), but for basic things like customer names or order details, a traditional database will be more efficient.

  • Running similarity searches can be computationally demanding. They may require more processing power and lead, potentially, to higher costs (depending on your cloud setup).

  • Compared to established databases, vector databases might have limitations in data management features like robust security or extensive transaction support.


Think of it like this: vector databases are amazing for understanding and connecting similar things, but they're not a one-size-fits-all solution. For simpler tasks or complex queries, traditional databases might be the way to go.

How are vector databases being used?


Vector databases are being used in a huge range of applications. These include things like search and recommendation engines, but it’s in AI where they’re having a particularly significant impact. This is because vector databases are used to integrate things called embeddings in AI applications — this helps an AI system to better ‘understand’ the relationships between the data on which it’s trained.

Doing AI? Do it the right way with Thoughtworks