Smart catalog search
Customers and staff search in natural language, for example by application or property. Search on a part number or NSN code stays exact: type a number and you get that one item.
AI on your own servers. Your product data never leaves your network.
Request an intakeFour use cases we start with. Each runs on your own infrastructure or on our dedicated bare-metal servers in the Netherlands.
Customers and staff search in natural language, for example by application or property. Search on a part number or NSN code stays exact: type a number and you get that one item.
The model reads datasheets and turns the specifications into structured product attributes and product pages. Every value points to its place in the source document, so a member of staff can check it.
Incoming emails with a quote request are turned into a structured quote request in the webshop or ERP: items, quantities and customer details. A member of staff checks and sends it.
Product content in Dutch, English and German, with a fixed glossary for your technical terms. The same term gets the same translation in each language, every time.
A question in plain language, an answer with the specifications next to it.
This is a static example. The company, the products, the part numbers and the NSN codes are fictional. No AI runs on this page.
We work with open-source language models such as Mistral, Llama and Qwen. The model runs on-premise at your site or on dedicated bare-metal servers in the Netherlands, with your catalog and documents as its source.
Because the model runs locally, no internet connection is needed to use it. The solution can therefore also run fully offline, in an air-gapped environment.
The starting point is that your data stays where it is. Product data, documents and customer requests are processed on the servers where the model runs and do not go to third-party AI providers.
We do not claim certifications and we give no legal guarantees. We do provide the technical information your security officer or data protection officer needs to assess the solution themselves.
Security & complianceFor companies with a large technical catalog and with data that must not end up with a third-party AI provider.
All servicesYou have thousands of items with specifications in datasheets, drawings and tables. That knowledge has to be findable for customers and for your own people.
You carry ranges from many suppliers, each with its own part numbers and documentation. Searching and quoting now take a lot of manual work.
You work with NSN codes and with requirements on where data is stored and who can reach it. A model that runs on-premise or fully offline fits that way of working.
Start with a proof of concept on your own catalog.
We discuss your catalog, your documents and the use case you expect the most from.
We build one use case on part of your own catalog, in your environment or on our servers in the Netherlands.
Your staff test it with real questions and documents. Together we measure what works and what does not yet.
If the result is good, we connect the use case to your webshop or ERP and extend it to the rest of the catalog.
Is your question about Private AI not listed? We always reply within one working day.
Ask your questionWe work with open-source models such as Mistral, Llama and Qwen. Which model we choose depends on the task, the languages and the available hardware. In the proof of concept we compare models on your own data. For each model we check that its licence fits your use.
That depends on the model and on the number of users. Language models usually need a server with one or more GPUs; catalog search needs less computing power. If you have no suitable hardware, the model can run on our dedicated bare-metal servers in the Netherlands. We determine the required capacity in the proof of concept.
Yes. The model and your data are on the same infrastructure, so no internet connection is needed to use it. The solution can run fully offline in an air-gapped environment. In that case you install updates of the model and the software through your own procedure.
You do. Your product data, documents and customer requests remain your property and are not sent to external AI APIs. We do not train models on your data. If you want a model tuned on your own data, that happens only on your instruction and the result remains yours. Logs and access are under your control.
That differs per use case and depends on the quality of your data. Search on a part number or NSN code is exact. With spec extraction from datasheets and with quote requests a model can make mistakes; that is why it shows the source for every value and a member of staff checks the result. In the proof of concept we measure accuracy on your own catalog, so you decide on the basis of figures.
Yes. For Magento 2 we build the integration as a module. We connect other webshops, existing catalogs and ERP systems through their API, or through file exchange if there is no API. Which integration is needed is determined in the intake.
Request an intake. We reply within one working day and discuss whether a proof of concept on your catalog makes sense.
Or call us: +31 485 841 000