Hardware
From the quiet compact system for the office to the redundant server pair for multiple locations. Delivered ready to run, set up by us. Or on your own hardware, after the hardware check.
AI solution
An AI system that belongs to you.
Not a subscription with a provider, but a device in your server room: hardware, open AI models, your company knowledge and the support that goes with it, from a single source and tailored to your company.
Four building blocks
What you get.
Every package consists of the same four building blocks, sized to fit your company.
AI models
Several vetted, open AI models, each doing what it does best. Not a “super AI” that does everything, but the combination of suitable models with your company knowledge. Your staff don't need to know which model is answering; we'll tell you if you ask.
Your knowledge
Documents, files, manuals and policies are connected, with roles and access permissions. Your AI answers from your sources and cites them.
Support
Setup, learning portal, live webinars, support with dedicated contact people and a customer portal that shows every step. People who know your system.
How an answer takes shape
From question to sourced answer in seven steps.
No single “super AI” does everything. The strength lies in the combination: suitable models, your company knowledge and your permission structure, brought together on your hardware.
- InputA question, a document or a task, via browser or app on your network
- ModelsThe AI selects the right model: text, specialist language, image or code
- Knowledge sourcesRelevant passages from your manuals, contracts, files and systems
- PermissionsOnly sources the person asking is authorised for under your permission structure
- ProcessingOn your hardware, on your premises. No connection to the outside
- AnswerIn your language, in your tone, with your terminology
- Sources & reasoningEvery answer names the documents it comes from, for review and verification
AI answers can be wrong. The source reference makes them verifiable, and responsibility for decisions stays with the people who make them.
Right-sized AI
The right AI.
Not the biggest possible AI.
The largest model is rarely the best for your task, but always the most expensive to run. We size the model and hardware to what you actually do.
01TaskWhat your staff do every day: quotes, contracts, enquiries, reports
02Suitable model classA different model for text than for images or specialist literature, each only as large as needed
03Suitable hardwareCompact system or server class, depending on users and models, not prestige
04ResultFast answers, quiet operation, a price that matches the task
Flow diagram: task, suitable model class, suitable hardware, result.
Individual AI customisation
What “tailored to your company” really means.
There are three levels, and we tell you openly which ones you need. Usually it's the first two.
Knowledge integration
Your documents are made searchable for the AI. No training required: a new manual is available immediately, a deleted one is gone immediately. Answers cite the source.
Configuration
Your terminology, your tone, your rules: who is allowed to see what, how the AI should respond, which capability handles which task. Configurable, traceable, changeable at any time.
Training
Genuine fine-tuning of a model on your data, but only when it actually makes a difference, for instance with very specific specialist language. We discuss this separately and honestly.
Connections to your systems
Your AI doesn't just know folders.
It works with your systems.
ERP, CRM, document management, Microsoft 365, Google Workspace, DATEV, databases: your AI can work with the systems your company already uses. In every connection it reads approved information and combines it. Where the interface provides for it and you approve it, it can also carry out defined actions: prepare a draft, create an issue, update an entry, hand data over.
- Runs on your network. The connector talks to your system, not to a cloud. The data stays where it is.
- Permissions stay permissions. Whatever an employee isn't allowed to see in the ERP, the AI won't show them either. Actions only run with the permissions you grant for them.
- Open standards. Under the hood MCP, APIs and OpenAPI: no proprietary format that ties you to us. New systems can be added later, as a custom connection where needed.
Connectable today
Always includedFile server, NAS, network drives
StandardMicrosoft 365, Google Workspace, email, Nextcloud, Confluence / Jira
Enterprise systemsERP (SAP Business One, Microsoft Dynamics, Sage, Odoo), CRM (HubSpot, Salesforce, Pipedrive), DMS (DocuWare, d.velop, ELO), accounting (DATEV, Lexware), SQL databases
CustomIn-house software, industry solutions, special interfaces
InformationRead, search, analyse, combine: in every connection
ActionsHand over, create, update, trigger: depending on interface and approval
Example: Sales asks, “What did we last quote to customer Müller?” Your AI answers from the CRM and the quotes folder, with a reference to both. The specific scope of functions depends on the respective system, the interface and your approvals.
Operation
Runs.
And stays under your control.
An AI system needs maintenance. How much of that you want to do yourself is up to you.
- Updates included: For twelve months we provide model and software updates. They're installed once you approve them.
- No internet connection required: The AI processing itself runs on your premises. A connection is only used for updates and remote service, and only when you open it. Connections to cloud services such as Microsoft 365 naturally use their own connection.
- Optional remote service: We monitor, update and maintain the system remotely, within the scope you approve, with a log.
- Expandable: Capabilities, users and performance level can be added later, with no need to buy new hardware.
An example
Friday, 4:40 pm. A customer asks about the warranty terms for a custom job from 2019.
Before, this meant searching the folder, calling a colleague and answering on Monday. Today, someone asks your AI and gets the answer with a reference to the contract and the handover record in seconds. The contract never left your premises.
Example scenario. Answer quality depends on the connected sources. For your IT team: technical architecture with data flow, permissions, updates and backup.
Future-proof
Your AI evolves with you.
You are not locking yourself into a single model. Your hardware stays, your company knowledge stays, and the AI can keep evolving: when a better model comes along, you replace the model, not your data basis. And whatever your company needs later, we add to the running system.
- Install new models: as an update, after your approval, without anything changing for your staff
- Add capabilities: Image & vision, Coding, specialist models, whenever the need arises
- Add users: a one-time fee per user, up to the limit of the hardware class
- Adjust roles and permissions: new departments, new access areas, changed structures
- Connect further knowledge sources: additional repositories, Microsoft 365, ERP, CRM, DMS
- Add integrations: via MCP, the open standard that doesn't tie you to any format
- Upgrade hardware: Performance tier + or a move to the next class, with your configuration and knowledge sources carried over
Today
Your own hardwareon your premises
Suitable AI modelmatched to the task
Your company knowledgesources, roles, permissions
Tomorrow
Existing hardwarestays
New AI modelis installed
Existing company knowledgestays
No new platform. No new data project.
Technically, a model change can require a new indexing run across your knowledge sources. Your source data stays within your infrastructure throughout. Model updates are included for twelve months, after that optionally as an update service.
Frequently asked questions
About the AI solution
Do I need to know about AI models?
No. You tell us what your AI should be able to do. We select and combine the right models. Day to day, your staff see one interface and ask questions. Which model answers is handled by the system.
What happens when a better model comes along?
The AI world moves fast. That's why your AI is built so that individual models can be swapped out without anything changing for your staff: you replace the model, not your data basis. Hardware, knowledge sources, roles and permissions stay. Technically, a model change can require a new indexing run across your knowledge sources; your source data stays within your infrastructure throughout. We provide these updates in the first year, after that optionally as an update service. The procedure is described in the technical architecture.
Does the AI work without an internet connection?
Yes. Processing of your requests runs entirely on the hardware on your premises. An internet connection is only used for updates and, if booked, for the remote service. Both happen only when you approve them.
How many employees can work at the same time?
That depends on the package and hardware. The packages are designed for 10, 50 and 150 users. If many people are processing large documents at the same time, we recommend the + performance level or the next larger package. The configurator shows you the price difference immediately.
Can I start small and expand later?
Yes, that's the normal case. Capabilities and users can be added at any time. If you switch hardware class, for example from a compact system to server class, we transfer your configuration and knowledge sources to the new system.
Which AI models are used?
Vetted, open models from AI research, a suitable one each for text, specialist language, images and code. We deliberately don't name specific models on the website, because they can change with every update; they are listed in your quote and in the technical description. What matters is not the name, but that the model fits your task and your hardware.
Can we connect our existing documents?
Yes, that's the core of it. File server, NAS and network drives are included in every package; Microsoft 365, Google Workspace, email, Nextcloud, Confluence as well as ERP, CRM, DMS and databases can be added as connections. The documents stay where they are. The AI reads them there and cites them as the source. Common formats such as PDF, Office files, scans and emails are processed; special formats we clarify in conversation.
Does DataFlowFacts develop its own AI models?
No, and we say that deliberately and openly. We rely on vetted, open models, select the right one for each task, combine them and adapt them to your company. Our work is the interplay, not the foundation model.
Ready for your own AI?
Put your package together in a few minutes, or first check whether your existing hardware is sufficient.