Production readiness#

Jina NOW supports multiple cross modal applications such as:

  • Text-to-text

  • Text-to-image

  • Image-to-text

  • Image-to-image

  • Text-to-video

  • Music-to-music

Jina NOW apps are production ready, our goal is to provide you with a search pipeline that is easy to integrate in your production system. We also come with other aspects such as AWS S3 Bucket data integration, security and finetuning.

Models we use:

  • SBERT : entence-transformers/msmarco-distilbert-base-v3

  • CLIP : openai/clip-vit-base-patch32

  • Music Encoder: BiModalMusicTextEncoderV2

For hardware we use AWS EC2 instances, from our experiments you can expect:

  • Queries per second: 13

  • Latency: 300 ms

AWS S3 bucket support#

We now support S3 Bucket, after selecting custom data you can select S3 Bucket. Then you can provide your AWS access key ID and AWS secret key, you can provide a list of emails of anyone you want to share your flow with.


Your data is automatically loaded from the S3 bucket. It will be temporary stored on EC2 machines while processing the data but it will not be stored.


You can choose to secure your Flow:


You can also choose to grant additional users access to the Flow:


You can do this by specifying their email addresses and seperate them using commas:



We also support finetuning, to enable this when you create your DocumentArray you also provide labels inside the tags more specifically the tag finetuner_label. This will automatically trigger the finetuning and your embeddings will be improved thus enhancing the quality of your final NOW search application.

More information on the training data format can be found here

Backend for Frontend#

We provide users with BFF (Backend For Frontend) API that you can call for your search application, which makes integrating to your custom frontend easy and seamless.


How to update your API keys#

To update your API keys use this simple script:

python now/admin/