Backup AI that runs on our own server

Kreaza's AI tools work with a cloud model. From now on, if it does not answer, short texts such as a product description and the store assistant's replies are written by a small model (Gemma 3, 4B size) that runs on Kreaza's own server, and the data does not leave it. We tested it on the server: a product description in Hebrew in about 5 seconds, and about 20 seconds the first time after a pause. The model's memory is capped so it does not hurt the rest of the system, and it is released after five minutes without use. Building a whole page or a long article still needs the cloud model.

New 1

  • A local AI model that handles short texts when the cloud model is unavailable.

Improved 1

  • Product descriptions and the store assistant keep working during an outage at the AI provider.