Generative AI can be much more than a public chat service. Businesses can operate their own AI environment in which employees work with different models, process internal documents and use AI to support recurring tasks.
Open WebUI provides a powerful, freely available interface for this purpose. It can be connected to different AI models and provides a shared environment for users, conversations and additional functions.
The key question is therefore not whether Open WebUI can be installed.
The key question is how the environment behind it is structured and who operates it.
How is an AI environment structured?
An AI application consists of several components that need to be distinguished.
Open WebUI is the interface that users work with. This is where they log in, have conversations, search their chat history and, depending on the configuration, can add files and documents.
Chat histories, user information and other data are stored on the designated server.
The actual AI model processes the submitted inputs and generates a response.
The process can be simplified as follows:
User
│
▼
Open WebUI
│
├── on a dedicated server
│ ├── within the company
│ └── at a suitable hosting provider
│
├── chat histories / files
│
▼
AI computer
│
├── within the company
└── at a suitable hosting provider
│
▼
Open WebUI
The server and AI computer can be located in the same place or operated at different locations. What matters is that the technical infrastructure and data flows remain transparent. The company decides which components are operated on its own premises and which are provided externally. I advise on the planning, selection and implementation of the appropriate solution.
Open WebUI is more than a chat window
Open WebUI can be understood, in simple terms, as a ChatGPT-like environment for a business.
Open WebUI itself is freely available. There are no subscriptions and no advertising. The software is operated on infrastructure selected by the company and connected to the desired AI models.
Employees can work with their chat histories, search conversations, add documents and – depending on the configuration – use different AI models.
Users and access rights can be managed within the environment. The functions and models available can be defined for each individual environment.
Key features at a glance:
- Chat histories
- Search within conversations
- Documents and files
- Different AI models
- Users and permissions
- Additional functions depending on the configuration, e.g. web search
This allows Open WebUI to serve as a central interface for generative AI within a company.
Try Open WebUI yourself
WebConcept One operates its own Open WebUI instance, where you can get a direct impression of the interface and its capabilities.
Practical support through AI
Local models do not necessarily match the capabilities of current frontier models. For the tasks mentioned, smaller AI models may already be sufficient and can be operated on comparatively affordable hardware. With more powerful hardware, larger local models can also be used, which are continuously being developed further.
For example:
- Drafting emails from bullet points
- Preparing and structuring information
- Summarizing longer texts
- Translation drafts
- Preparing recurring communications
- Assistance with research in internal documents
- Reviewing texts and providing initial indications of inconsistencies
For example, a language service can use a local model to create an initial translation draft. The professional and linguistic review remains the responsibility of the employee.
AI does not replace the actual service. It can take over part of the preparatory work.
For tasks like these, local models can be particularly interesting when the content being processed should not leave the company.
Data protection begins with the technical architecture
Data protection in an AI environment cannot be addressed with a single label alone.
What matters is which data is actually processed and which systems are involved.
This includes, among other things:
- Stored chat histories
- User accounts
- Uploaded files and documents
- Data transmitted to an AI model
- Logging and administrative data
- Backups and their storage locations
When local models are used, processing can take place within the company’s own infrastructure.
If an AI environment runs on infrastructure operated by WebConcept One, the respective responsibilities and access rights are defined transparently. Where data processing on behalf of a customer applies, this is regulated contractually, for example through a data processing agreement.
Transparency about the infrastructure is the basis for a comprehensible approach to the processed data.
The starting point is not the statement “everything is GDPR-compliant”, but the question:
What exactly happens to which data, at which point – and who can access what?
Which models and how much computing power?
Different models can be used for AI processing. Which models are suitable depends, among other things, on the tasks, the desired quality and the computing power required.
Alongside powerful proprietary models from major providers, there are open-weight models that – depending on the model, licence and required computing power – can also be operated on dedicated hardware.
Local models do not necessarily match the capabilities of current frontier models. For many practical tasks, however, the maximum capabilities of a model may not be necessary.
For example, a smaller model can create an email draft from bullet points, summarize a text, structure information or prepare a translation draft.
The key consideration is therefore which model delivers sufficiently good results for the specific task and how much computing power is required for it.
I advise on the selection of suitable models, test them for the intended tasks and assess the AI computing power required.
The necessary computing power can be provided in different ways. It can be provided by WebConcept One or built within the company.
If an in-house AI infrastructure makes sense, I support the selection, sizing and setup of the AI computer.
Models can be tested and compared with one another. New models can be added later or existing models replaced.
Model selection is therefore not a one-time decision that determines the entire environment.
What the service can include
The specific scope of services depends on the desired architecture and the requirements of the company.
Possible services include:
- Planning the technical architecture
- Installation and configuration of Open WebUI
- Selection and testing of suitable AI models
- Setting up users and access rights
- Configuring storage and data management
- Setting up an AI computer
- Connecting and configuring AI computing resources
- Configuring backups
- Documenting the environment
- Training employees and administrators
- Support with ongoing administration
Which components are operated by the company itself and which tasks are to be handled externally can be determined individually.
Operate it yourself or have it managed
An in-house AI infrastructure does not have to be operated permanently by an external service provider.
After the technical setup, the environment is documented, including its structure and the tasks required for ongoing operation.
These can include, for example:
- Updating Open WebUI
- Operating system and security updates
- Monitoring available storage capacity
- User and access rights management
- Backups and recovery
- Updating and replacing AI models
- Maintenance and expansion of AI hardware
The necessary administration can then be handled internally or supported as required.
The service therefore does not necessarily end with the installation. It can also consist of planning, training, documentation and handing over an environment that is fully understood by the company.
An individual AI environment tailored to specific requirements
Which infrastructure makes sense depends on the requirements of the company. I advise you on the planning, clarify the technical requirements and implement the chosen environment together with you.
If an AI infrastructure is to be operated within the company, existing systems can be incorporated or servers and AI hardware can be built and configured to match the requirements. The necessary infrastructure can also be provided externally.
The solution remains expandable after implementation. Models, computing power and other components can be adapted as requirements change.
Would you like to find out what such an AI environment could look like in your company? In a non-binding initial consultation, we will clarify your requirements and the possible approaches to implementation.