Private ChatGPT for business: four operating models compared

Introduce a controlled AI workplace for employees? AI Cube Pro provides the local AI chat with hardware, Open WebUI, an agreed model, hardening and initial setup. For automatically connected company knowledge, WZ-IT builds an internal AI assistant with citations and enforced permissions on top. Discuss the right starting point

People searching for a “private ChatGPT for business” rarely mean an exact copy of OpenAI's product. They are looking for a central AI workplace where employees can chat, analyse files, use approved models and access internal knowledge. The decisive questions are therefore not limited to the user interface. They include who operates the platform and models, where data travels and how company knowledge and permissions are integrated.

This guide compares four operating models and the questions to answer before choosing one. As of August 2026.

Table of contents

• What businesses mean by private ChatGPT
• Four operating models compared
• What employees actually need
• Company documents and internal knowledge
• Data protection and controlled data flows
• Compare costs fairly
• Which option fits
• A local start with AI Cube Pro
• Selection checklist
• Sources

What businesses mean by private ChatGPT

ChatGPT itself is a cloud service operated by OpenAI. Its closed models cannot be installed on a company server. A business can, however, build a functionally comparable AI workplace from open or externally connected models, a chat interface, identity management and controlled knowledge sources.

From an employee's perspective, that normally includes:

• a familiar browser-based chat interface;
• approved models for text, files and possibly images or speech;
• personal and shared knowledge collections;
• users, groups and distinct model or data permissions;
• traceable citations for answers grounded in company documents;
• central administration, updates and support.

The phrase “private ChatGPT” therefore describes the user experience. The underlying system may be an external workspace, a dedicated platform or a local AI server.

Four operating models compared

No option is right for every organisation. “Hosted in the EU” does not automatically mean that every model request remains within the EU. A local installation is not automatically secure or GDPR-compliant either. The complete architecture determines the result.

What employees actually need

Model quality is only one part of the product. An internal AI chat becomes useful when login, permissions, models, knowledge and operations work together.

A workspace such as Open WebUI can expose local models and deliberately approved external endpoints through one interface. Employees can chat, work with files, configure assistants and use personal or shared knowledge collections. OIDC can connect authentication to an existing identity provider.

Before rollout, organisations should decide:

• Which groups may use which models?
• May external models be available alongside local models?
• Which data may enter personal or shared knowledge collections?
• How are leavers, group changes and retention periods handled?
• Who owns updates, model changes, backup and support?

Company documents and internal knowledge

For an initial rollout, employees can add individual files or curated collections to the workspace. That differs from a continuously synchronised company search.

Connecting Nextcloud, SharePoint, a DMS, file shares, wikis, tickets or databases automatically requires a RAG pipeline. It ingests approved content, prepares it for retrieval and passes relevant sections to the model as context. A sound implementation identifies the document, section and version behind each answer.

Permissions must apply before retrieval. A user must not receive a passage merely because it exists in a shared index. This is the purpose of WZ-IT's internal AI assistant for company documents. The guide to RAG with Nextcloud, SharePoint and DMS explains the source layer.

Data protection and controlled data flows

OpenAI states that business data is not used to train its models by default. A business assessment must also cover the contract, plan, retention, subprocessors, region, enabled features and support access.

A local platform can keep model inference inside the company network. Updates, telemetry, web search, external models, email, backups or remote support may still create outbound connections. These paths need to be documented and technically controlled.

At minimum, assess:

• purpose and permitted data categories;
• legal basis and any data processing agreement;
• user, group and administrative access;
• logging, retention and deletion;
• subprocessors and international transfers;
• risk assessment and any required DPIA;
• employee rules for acceptable input.

The GDPR-compliant AI checklist covers these controls in detail. Patient, client and other professional secrets also require the contractual and technical controls described in local AI for confidentiality professions.

Compare costs fairly

Cloud workspaces often charge per user or by consumption. A local platform moves cost into hardware, setup and operations. A fair comparison uses the same functional scope across at least three years:

• user licences or model consumption;
• hardware, power and replacement risk;
• platform, identity and network integration;
• knowledge sources and the RAG pipeline;
• updates, monitoring, backup and support;
• migration and exit.

A cloud workspace may be more economical for a small number of users and general AI chat. Stable usage, sensitive data or deep knowledge integration can favour the control and predictable capacity of a dedicated platform. What does a local AI server cost? provides a practical calculation framework.

Which option fits

A cloud workspace tends to fit when a team needs general AI quickly, does not want to operate a platform and may process the intended data within the reviewed contractual framework.

A dedicated or European platform tends to fit when operations should remain outsourced but region, models, access and exit need stronger control.

A local platform tends to fit when inference and knowledge should remain within the company network, offline operation matters, models should remain selectable or sensitive information is involved.

A hybrid design tends to fit when local models handle confidential tasks while approved cloud models remain available for other work. Model choice and data classification then need to be visible to users and enforced technically.

A local start with AI Cube Pro

AI Cube Pro is WZ-IT's preconfigured starting point. Hardware, Open WebUI, a local model runtime and an agreed model are prepared, hardened and tested. Initial setup and support are included. After the configuration has been agreed, the system is generally ready for use inside the business within two weeks.

This first provides a local AI chat for employees. Personal and shared knowledge collections can be created directly in Open WebUI. Automated sources, SSO, special permission architectures, business-system interfaces and custom RAG pipelines are then added around the existing environment.

For larger models, more concurrent users or special availability requirements, WZ-IT designs AI Cube Custom, additional AI Cubes or dedicated GPU servers. The company can operate the platform itself or hand it over as managed AI.

Selection checklist

1. Which three tasks should employees complete with the system?
2. Which data classes may be processed locally, in the EU or by an external provider?
3. Is a general chat sufficient, or must company knowledge be connected automatically?
4. Which user groups and document permissions must be enforced?
5. Which models meet the required quality, language and speed?
6. How many answers are likely to be generated concurrently?
7. Who owns updates, monitoring, backup and support?
8. How can data, configuration and models be exported during an exit?

Only these answers determine whether a cloud workspace, a dedicated platform, AI Cube Pro or a custom deployment is the appropriate route.

Sources

• OpenAI: enterprise privacy and business data
• OpenAI: ChatGPT Business
• Open WebUI: self-hosted platform documentation
• Open WebUI: Knowledge
• European Commission: GDPR obligations for organisations
• WZ-IT: Local AI or ChatGPT Business?