Skip to content
MYNA
ENDE
All guides

Local AI or cloud AI: what fits your business?

Review contracts, analyse documents and transcribe meetings: local AI handles these tasks without sending your content to a cloud AI provider.

MYNA · 17 October 2026 · Sources and provider statements checked

Local or cloud: Where does your data go?

Local AI processes your questions and files on a computer in your company. Cloud AI sends this content over the internet to a provider, where the answer is generated.

A locally run language model is also called a local LLM. It can summarise text, answer questions about documents and write drafts. A cloud server in Switzerland is still an external service.

What the data flow looks like

Local

  1. Your questionWith approved data
  2. Your AI computerProcessing on your premises
  3. Your answerBack to your team

Cloud

  1. Your questionOver the internet
  2. External providerProcessing by the service
  3. Your answerBack to your team
Simplified data flow: processing within your business or at a cloud provider.

The benefits of local AI

  • Confidential content stays in your company. Analyse contracts, cost calculations and internal reports without sending them to a cloud AI provider.
  • You control access. Your company decides who may use each document and its results.
  • No upload for someone else’s model training. Local processing keeps your inputs away from cloud providers.
  • More control over operation. You decide when models are updated and stored content is deleted.

This requires suitable hardware, setup and maintenance. MYNA sets up the solution and matches the model and computing capacity to your tasks.

Cloud risks: Your SME remains responsible

When you give personal data to a cloud service, your company remains responsible for its lawful processing. This includes customer lists, job applications and personnel files. FDPIC: Responsibility for cloud services (DE)

  • Additional recipients: The provider may use subcontractors. You need to know who processes your data and in which countries.
  • Limited control: Standard contracts often leave little room to negotiate storage, deletion or your own requirements.
  • Transfers abroad: Check the conditions for the countries involved before uploading.

The FDPIC sets out these points in its cloud guidance (DE). The revised Federal Act on Data Protection (FADP) permits cloud use subject to conditions, notably Article 9 on processors and Articles 16 onwards on transfers abroad. FADP on Fedlex

What to keep out of an unapproved cloud chat

  • Credentials: Replace passwords and API keys with placeholders.
  • Contractually protected content: If a customer contract forbids sharing code or plans with external services, that also applies to a paid AI account.
  • Patient and client files: Check professional secrecy and permitted processing before using an external service.
  • Personnel records: Do not copy medical certificates, salary details or job applications into personal AI accounts.

The Swiss NCSC warns against inadvertently sharing business secrets and personal data with AI tools. NCSC: AI at work (DE)

There is no blanket cloud ban for health data or other sensitive personal data. A service must meet the applicable legal and contractual requirements. Canton of Zurich: Professional secrecy and cloud services (DE)

Cloud licences: Check training and storage

A personal paid subscription does not automatically offer the data protection of a business account. The specific licence, contract and settings matter.

If training on your inputs is enabled, explicitly switch it off before business use. Some business services already exclude it. Without a reliable training exclusion, do not use the service for confidential business data. NCSC: Business accounts and opt-out (DE)

OpenAI API

OpenAI says API data is not used for model training by default; customers can explicitly opt in. This does not mean zero retention: abuse-monitoring content may normally be kept for up to 30 days, or longer in certain cases. Features have additional storage rules. OpenAI: API data controls

Microsoft Copilot in a business account

For the documented Microsoft 365 commercial services, Microsoft excludes foundation-model training on prompts, responses and Graph data. Content is still processed and may be stored. Check your specific licence and additional services. Microsoft: processing and training

“No training” does not mean “no storage”. Check both separately. A chat instruction such as “Do not use this for training” replaces neither contractual terms nor account settings.

Which solution fits each task?

  • Public text: An approved cloud service may be enough to draft a product description.
  • Confidential documents: For customer contracts, internal cost calculations or personnel files, local AI offers a clear benefit: the content stays with you.
  • Internal meetings: Local transcription produces text and draft minutes without sending the recording to an external AI service.

How MYNA implements your solution locally

MYNA can carry out the tasks described here entirely locally: working with documents, answering questions and transcribing audio. Your data stays within your business. We tailor the feature set to your local operation.

How to get started with local AI

Choose a recurring task, such as answering questions about customer contracts or preparing meeting minutes. Test MYNA with a document or recording you know well. Check whether the result is correct and reduces your workload.

In a free introductory call, we discuss the task, the data it needs and the right setup.

Further reading

Legal and provider information checked on 17 October 2026. Rules may change; applicable law and your specific agreement govern.

Let’s look at your task

In a free 20-minute introduction, we explore whether and how MYNA can help you.

Meet usExplore MYNA Solution