---
title: Choose AI. Know where your data goes | Karbonic
canonical: https://www.karbonic.net/donnees-ia-en.html
meta-description: 'Local AI, hosted services or a combination: quality, costs, confidentiality
  and data recovery are examined together.'
description: 'Local AI, hosted services or a combination: quality, costs, confidentiality
  and data recovery are examined together.'
language: en
og:title: Choose AI. Know where your data goes | Karbonic
og:description: 'Local AI, hosted services or a combination: quality, costs, confidentiality
  and data recovery are examined together.'
og:image: https://www.karbonic.net/assets/orchestration-v3.webp
og:locale: en_GB
og:type: website
og:url: https://www.karbonic.net/donnees-ia-en.html
twitter:card: summary_large_image
twitter:title: Choose AI. Know where your data goes | Karbonic
twitter:description: 'Local AI, hosted services or a combination: quality, costs,
  confidentiality and data recovery are examined together.'
twitter:image: https://www.karbonic.net/assets/orchestration-v3.webp
theme-color: '#17212c'
---

Source: https://www.karbonic.net/donnees-ia-en.html

Data and AI

# Choose AI. Know where your data goes.

Local AI, hosted services or a combination: quality, costs, confidentiality and data recovery are examined together.

![Abstract illustration of organised information flows.](https://www.karbonic.net/assets/orchestration-v3.webp)

## Choose AI based on your work

Reading files, preparing texts or classifying documents: an AI tool can be tested against a specific need. Result quality, necessary corrections, cost and confidentiality are considered together.

Choosing AI is part of organising your IT. Files, applications and backups can stay on your equipment, be hosted or be distributed across environments.

[Hosting and data control](https://www.karbonic.net/hebergement-donnees-en.html)

## Test on a file before extending the approach

**Example need.** An engineering office wants to retrieve information from its files without sending documents to an external service. A representative sample is used to test a local model, document search and source checks.

The local runtime, OCR if needed, document index and file access are examined together. Each expected answer is compared with the original records. Speed, errors and information that cannot be found are recorded before deciding on hardware or wider deployment.

Karbonic also develops a document synthesis engine that links each retained fact to a source and applies explicit checks. This work forms part of information-processing projects whose scope is defined around the business need.

## Know where your information travels

Documents sent, retained data, backups and support access are identified. You can understand who receives your information and under what conditions. The stated server country does not answer all these questions.

## Compare the options

| Option | What it enables | What to plan |
| --- | --- | --- |
| On your equipment | Run compatible models in your environment. | Suitable hardware, updates, backups and control of external connections. |
| --- | --- | --- |
| With a hosted service | Use a service without managing all its infrastructure. | Contract, data access, retention, usage costs and file recovery. |
| --- | --- | --- |
| Combining both | Keep some processing on your equipment and use a service for the rest. | Separation rules and checks on transferred information. |
| --- | --- | --- |

## Local AI in practice

A local runtime runs compatible models on your equipment. Connections to external services are disabled or limited according to the agreed scope. Connected applications, extensions and backups are also examined to establish what actually leaves your environment.

Hardware is chosen after a representative trial. Document volume and required speed matter as much as the model name.

## Confidential information

Access, subcontractors and transfers are described. For a law firm or another sensitive activity, this technical analysis supports the examination of confidentiality obligations and case-specific legal requirements.

## Be able to change tools

Exports and backups are planned from the outset. A test checks that documents, attachments and associated information can be recovered and read elsewhere.

[Swiss data protection law, GDPR and CLOUD Act: practical guidanceFR](https://www.karbonic.net/journal-donnees-cloud.html)

A concrete starting point

## Let's discuss what you want to improve.

Describe the difficulty. Options and the work required will be discussed with you.

[Let's discuss your situationFR](https://www.karbonic.net/contact.html) [Work, budget and schedule](https://www.karbonic.net/accompagnement-en.html)