Skip to content
Platform

Data Processing

On the Evoya platform, you can control where your data is processed. The processing location depends on the respective feature – text, images, embeddings, dictation, Konversa, or documents. For most features, administrators can configure the location themselves; for some, it is determined by the product or model.

This page covers only the processing of data. Where data is stored is described on the Data Storage page.

Overview

Feature Processing location Configurable?
Text (LLM) Switzerland, Europe, USA, or Other regions – depending on the selected model Yes, per agent or as a standard model
Vision (Image-to-Text) Switzerland No, Swiss model by default
Documents (OCR) Switzerland No, Swiss model by default
Embeddings (data sources) Switzerland or USA Yes, standard region and per data source
Speech-to-Text (chat dictation) Switzerland No, Swiss model by default
Konversa Switzerland or EU – depending on product licence Determined by product licence
Image generation USA No, only US models available

Text (Large Language Models)

For text-based AI features, the processing location depends on the selected Large Language Model (LLM). The models are grouped by region:

  • Switzerland
  • Europe
  • USA
  • Other regions (e.g. China)

Which model is used is defined when configuring the respective agent. At the organisation level, administrators can also define a standard language model that is preselected when creating a new agent. This standard model can be overridden per agent.

→ Learn more: Introduction to AI Models, Manage models

Vision (Image-to-Text)

When extracting information from images (Image-to-Text), the platform uses a Swiss model by default. Processing therefore takes place in Switzerland.

This location is currently fixed. Soon, it will be possible to choose between Switzerland and EU.

Documents (OCR)

When content needs to be extracted from scanned documents or images, OCR. To additionally improve the results, the platform uses a vision model – a Swiss model by default. Processing therefore takes place in Switzerland.

This location is currently fixed. Soon, it will be possible to choose between Switzerland and EU.

Embeddings (data sources)

When data sources are indexed, you can specify where the data is vectorised. The available options are:

  • Switzerland
  • USA

Administrators can define a standard embedding region on the Manage models page, which is preselected for new data sources. By default, this is set to USA and can be changed to Switzerland if needed. The region can be overridden per data source.

→ Learn more: Data Sources

Speech-to-Text (chat dictation)

When speech is converted directly into text in the chat, the platform uses a Swiss model by default. Processing takes place in Switzerland.

Konversa

For Konversa, the processing location depends on the product licence your organisation holds:

  • Konversa CH – processing in Switzerland
  • Konversa EU – processing in the EU

Which variant is available depends on the product you are licensed for. The location cannot be freely configured but results from the licensed product variant.

→ Learn more: Product Access

Image generation

For image generation, only US models are available. Processing therefore takes place in the USA.

Selecting other regions is currently not possible for image generation.