Instead of relying on a single model, Insiders combines external and proprietary large language models (LLMs) into a single solution for your document workflows. The way this solution handles data remains transparent and compliant with data protection regulations.
How many of your document processes still rely on someone correctly interpreting free-form text? Large language models understand, process, and generate natural language. That’s exactly why we use them for classification, extraction, and automation in document processing—they’re firmly integrated into our products.
A Large Language Model, or LLM for short, is an AI model that has been trained on very large amounts of text and has learned how language is structured. It recognizes patterns in phrasing, categorizes terms, and deduces what a text is about—without anyone having to provide it with fixed rules.
For document processing, this is the key difference from traditional text recognition. A rule-based system requires a separate configuration for each invoice template. A language model, on the other hand, can understand even a document it has never seen in this form before—a complaint written in free-form language, an invoice with an unfamiliar layout, or a termination notice without a subject line.
What matters here is not whether a model understands language. What matters is which model is used for which task, where it is deployed, and what happens to the data it processes.
Compatible Model
Privacy Policy
Legally sound
Knowledge and Action
A single language model cannot handle every task equally well. And relying on a single provider quickly becomes a risk, both technically and contractually: prices change, models are discontinued, and interfaces break. That’s why OvAItion combines external and in-house LLMs and AI agents trained for document processing, depending on the use case.
Extracting individual fields from an invoice doesn’t require a model that can write poetry. Conversely, a highly specialized extraction model is of little help when a customer complaint written in natural language needs to be categorized. The platform decides which model handles a task—not you.
AI in business boosts efficiency, reduces costs, and unlocks new opportunities for innovation in nearly every industry. Our customers’ use cases demonstrate how successful process optimization makes companies more efficient and innovative. By implementing intelligent solutions, they have been able to optimize their business processes and achieve sustainable competitive advantages. Let our success stories inspire you!
AI agents are increasingly planning, deciding, and acting independently – yet without central control, traceability is lost. OmnIA integrates agentic AI applications into your existing IT landscape in a controlled, explainable, and data-protection-compliant manner.
External models are available immediately and perform well on general language tasks. Private models run on your own infrastructure and never leave it. For most document-processing tasks, the right answer isn't an either/or decision, but rather a matter of categorizing data based on its sensitivity.
| Criterion | External LLM | On-Premises Private LLM |
|---|---|---|
| Where data is processed | At the model provider’s location | In your data center or the Insiders Cloud (ISO 27001, BSI C5 Type 2, located in Germany) |
| Setup effort | Minimal, ready to use immediately | Higher, but a one-time cost |
| Ongoing costs | Per request; increases with volume | [PLEASE FILL IN: Server requirements and the volume at which in-house operation becomes cost-effective] |
| Dependence on providers | Price and model changes take effect immediately | None—the model and version remain under your control |
| Suitability | General language tasks, free-form expressions | Recurring document processes involving sensitive content |
The model market is changing faster than selection decisions become outdated. That’s why we regularly evaluate the available language models based on what matters most in document processing: recognition accuracy, processing speed, and cost per document.
We publish these results openly—most recently in December 2025. They are directly incorporated into the model selection process in OvAItion. If a new model performs better for a given task, you benefit from it without having to make any changes yourself.
