AI that never leaves your premises.
HR data, cost calculations, customer files: some things do not belong in a cloud. With local language models, the AI runs on a server at your premises, and webx cortex makes sure that sensitive requests do not go outside in the first place.
- Models on your server
- Privacy level per assistant
- No training on your data
- Cloud only where you allow it
Where a request goes
As local as your data requires.
Not every task needs the same level of caution. webx cortex determines per task which model is allowed to work.
Cloud (EU)
Powerful cloud models via EU endpoints with a data processing agreement (DPA), for non-critical tasks such as texts and translations. Personal data is masked beforehand.
Hybrid by rules
Every assistant is given a privacy level. With "local only", requests go exclusively to local models, all others may use the cloud.
Fully local
Language model, knowledge database and search run on a server at your premises. Your data does not leave your network.
What local models can do, and what they cannot.
Well suited
Summarising, classifying, answering questions about your documents, extracting data from texts, writing drafts. For many everyday tasks, a local model is enough.
Better in the cloud
Long, nested analyses and demanding texts are usually handled better by large cloud models. That is why the LLM router separates by task: local where sensitive data is involved, cloud where performance is what matters.
What you need for it.
For the search
Capturing and searching your documents (embeddings) runs on a normal server without a graphics card.
For the language model
A local language model only responds at a usable speed with a suitable graphics card. Which size fits, we calculate in the workshop with your use cases before you buy anything.
The same security levels, wherever the model runs.
PII scrubber
Names, email addresses and phone numbers are detected and masked before every model call.
Prompt injection protection
Inputs and outputs are filtered so that manipulated documents or emails cannot give the AI false instructions.
Log
Every model call is logged, including the question and which model answered.
More from webx cortex.
AI knowledge management
Ask for company knowledge instead of searching for it, with source references and permissions.
AI funding Tyrol
Data sovereignty and AI integration are eligible for funding, the application must be submitted before the project starts.
Frequently asked questions
What is local AI?
Local AI means that the language model does not run at a cloud provider, but on a server in your own network. Requests and data do not leave your premises. Freely available open-source models are used.
Is a local model as good as ChatGPT?
Not in everything. Large cloud models are usually stronger for complex analyses and demanding texts. For summarising, classifying and questions about your documents, a local model is often enough. webx cortex can combine both: local for sensitive data, cloud for the rest.
Do we need our own hardware?
For fully local operation, yes: a server with a suitable graphics card at your premises or data centre. The size depends on the use cases, which is why we calculate it in the workshop before you buy anything. Hardware is a third-party cost and is shown separately in the quote.
What happens if the local model fails?
Assistants with the "local only" level then do not answer. They deliberately do not fall back to a cloud model, so that sensitive data stays on your premises even in the event of a fault.
Which data should stay with you?
We look at which tasks have to run locally and which may use the cloud, and tell you what that means in terms of hardware. The initial consultation is free of charge.
or by email to info@web-crossing.com · To the AI overview