Vector databases

A vector database serves two main purposes. First, it stores the values in the previously generated embeddings. Second, it provides a way to query the values. When searching the store you choose how similarity is calculated. Common algorithms are Cosine similarity, Dot product, and Euclidean distance.

Each vector database in the ecosystem supports a different list of algorithms. The Vectorize team has analyzed each vector database (and its supported algorithms) in the ecosystem to offer the best stores to you.

The data that you provided in your experiment was first chunked and then an embedding was generated for each chunk. Then the vectors within the embedding and the original chunk of data are stored in the chosen vector database.

While running an experiment, the Vectorize platform formulates test questions and then queries the vector database using those questions. The average relevancy score is an index showing you how accurate the result of the queries was to the original question.

All the supported vector databases use the Cosine similarity algorithm when running experiments.

Supported vector databases

The Vectorize platform offers integrations with the following vector databases. Click to learn more.

Last updated