netAI is the AI model family from Net-Base. Instead of one large cloud AI for everything, we develop small, specialized models that run entirely locally in our applications – without a cloud account, without transmitting your content, and without ongoing AI costs.
miniAISpelling, phrasing and text structure – local, for PCs with a GPU or a powerful CPUIn development
Further modelsEach for a clearly defined task in a Net-Base applicationPlanned
LocalThe models run on your computer, not in the cloud
SpecializedEach model handles a few tasks — and does them correctly
HonestReports when it is uncertain, and does not invent things
Trained in-houseOwn servers instead of external AI services
The model family
A family in tiers.
Each tier is built for different machines and tasks. Common to all: they run on your machines, are specialized for their task and originate entirely from Net-Base.
Available in netPDF
microAI
Runs on any PC directly on the CPU, without a graphics card. It answers questions about netPDF, PDF, e-invoicing and printing, reads invoices, converts them into ZUGFeRD or XRechnung and learns from your confirmations.
The next stage assists with spelling, phrasing and structuring texts. It is intended for PCs with a graphics card or a powerful CPU and also runs fully locally.
Upcoming
Additional models
Further specialized models are being developed for individual tasks in Net-Base-programs. We will present them as soon as they are finished.
Principles
What netAI stands for.
Six rules apply to every model of the family – from the initial design to the update on your computer.
Local rather than cloud
The models run in the Net-Base programs on your computer. They require neither an internet connection nor a cloud account to operate.
Specialized rather than general-purpose
Each model is built and trained for a small set of clearly defined tasks. That makes it compact, fast and reliable for its intended purpose.
Honest
If a model is uncertain, it says so and flags the passage. It does not invent information or fill gaps with conjecture.
Learns from your confirmations
What you review and confirm is incorporated into the model on your computer. In this way it improves with each use.
In-house development
Models, learning procedures and runtime are developed by Net-Base itself. No external AI service is involved and there is no dependency on major AI providers.
Privacy by design
The contents of your documents do not leave the computer. This is not a setting but an architectural decision – and makes deployment GDPR-friendly.
Data protection
Your content stays with you. Our training remains in-house.
Data protection at netAI is not an add-on but the starting point. Models are trained at Net-Base on dedicated servers and run locally on your systems. There is no cloud AI provider in between that can access the data.
Training in-house – on dedicated servers at Net-Base, not with a cloud AI provider
Training data stays on-premises – it does not leave Net-Base
Models without texts – only weights of hashed features enter the models, no texts, names or amounts
Local only – documents, invoices and queries do not leave your computer; learned information is stored encrypted in your user profile
Sharing without content – sharing learned data transmits only weight changes and document templates without contents, and can be disabled at any time
1
The teacher stays in-house
A large teacher model runs only on servers at Net-Base. It trains the small models and verifies that they perform their task sufficiently well.
2
Small student models learn
From this, small, specialized student models are created for each specific task—so small that they run quickly on standard PCs.
3
Only the students come to you
Program updates deliver only the small models to you. The teacher model and the training data never leave the premises.
Frequently asked questions
netAI explained briefly.
What is netAI?
netAI is the family of proprietary AI models from Net-Base. Each model is small, specialized for a clearly defined task and runs entirely locally within a Net-Base program. The first model is microAI in netPDF; further tiers such as miniAI are in development.
Where are the models trained?
At Net-Base on-premises, on dedicated servers—not with a cloud AI provider. Training data does not leave the premises. Only weights of hashed features enter the models, no texts. The large teacher model also remains on-premises; only the small, specialized student models come to you.
Are my documents used for training?
No. Your documents, invoices and queries do not leave your computer. The models learn locally from your confirmations. If „Share learned data“ is enabled in netPDF, only weight changes and document templates without contents are transmitted—the sharing can be disabled at any time.
What is miniAI?
miniAI is the next stage of the family and still under development. It is intended to assist with spelling, phrasing and the structure of texts, is aimed at PCs with a graphics card or a powerful processor, and also runs entirely locally. We will announce it as soon as it is finished.
Do I need a graphics card for netAI?
Not for microAI: it runs on any PC directly on the processor. Only the upcoming stage miniAI is intended for machines with a graphics card or a powerful processor.
Be part of it today
Try microAI in netPDF.
The first model of the family is integrated into the PDF editor netPDF: press F1 and ask in your own words, or have a PDF invoice converted into an e-invoice. All features for 30 days after free registration.