Embedding Similarity Search

We enable you to integrate embedding similarity search capabilities into your products and applications.

Quick start

Required steps:

  • Create a dataset
  • Import your embeddings to our API
  • Build the index and wait till it's done
  • Set embedding length
  • Send requests to our API

Create a Dataset

Create a dataset and choose embedding-similarity-search as dataset type. The dataset contains all information about your index.

API reference
https://api-reference.ioannotator.com/#operation/createDatasets

Import your embedding

When importing your embeddings you have to provide a embedding, dataset and a referece which is used for you to reference to your specific results.

API reference
https://api-reference.ioannotator.com/#operation/importEmbedding

Embedding length

The index is built for your specific embedding length that's why we need to know that before you can build the index. We support any arbitrary embedding length.

API reference
https://api-reference.ioannotator.com/#operation/setEmbeddingLength

Similarity vector size
Similarity vector size

Build index

The index can be build at any time. There is no need to wait till all embeddings are imported (Just rebuild at any required time). Depending on the amount of embeddings the index rebuild might take a while.

API reference
https://api-reference.ioannotator.com/#operation/rebuildIndex

You can check if the index build is completed

API reference
documented soon

Send search requests

You can send search requests against your specific endpoint as soon as the index build is completed.

API reference
documented soon

FAQ

Can I delete already imported embeddings?

This is currently not possible. The feature is on our roadmap. We will give you soon the possibiltiy to remove embeddings using the reference.

The search endpoint is returning 401 unauthenticated

Please check if the API key you are using is the same as configured in our API key management https://app.ioannotator.com/api.

Can I import embeddings of different length into the same dataset?

No