SWPanel AI engine
Our AI, in our Data Centers
SWPanel's AI engine runs on our own servers, in our own Data Centers within the European Union, and processes your data under GDPR. Here is how it works inside: where it lives, how it isolates each customer's data, how it builds each answer and how much it uses.
- Own Data Centers in the EU
- No third-party AI services
- SWPanel calculates, the AI writes
-
1
Your data, in the EU Services, metrics and logs, in our Data Centers.
-
2
SWPanel calculates The figures, rules and permissions of whoever asks.
-
3
A single gateway Each AI job has its own engine, model and limits.
-
4
Own GPUs NVIDIA H100, A100 and RTX 6000, on Dell PowerEdge servers.
-
5
Validation If the text contradicts the figures, it is redone.
-
6
Your answer Written in your language and with what backs it up.
How it works inside
From your question to the answer, on our own hardware
SWPanel's AI engine doesn't rent intelligence from anyone. It runs on our Dell PowerEdge servers with NVIDIA GPUs, in our European Union Data Centers. Each request goes through SWPanel, a single gateway and several engines that pass the work along, and only comes back when it matches the figures. Your data doesn't go to any AI provider.
-
1
A request arrives From chat, a nightly consultant, a report or a translation. Heavy work goes into a queue and runs at night.
-
2
SWPanel prepares the context Checks that the service belongs to whoever asks, calculates the figures from their data and passes along only what's needed.
-
3
The gateway distributes Each job has its own engine, token cap and time limit. The gateway measures each call and doesn't store the conversation.
-
4
The engines pass the work along The question becomes vectors, the index finds the fragments, they are reranked and judged, and the language engine writes with them and with SWPanel's figures.
-
5
It is validated If the text contradicts a SWPanel figure, it is requested again. The model doesn't decide any number.
-
6
Your answer comes back In your language and with what backs it up. Reports include their energy and CO2 accounting.
Servers
- Dell PowerEdge R760 Several units
- Dell PowerEdge R750 Several units
NVIDIA GPU
- H100 80 GB of memory
- A100 80 GB of memory
- RTX 6000 96 GB of memory
A single entry point
Each job, with its engine and its limits
Everything SWPanel asks of the AI goes through the same gateway. You don't ask “a model”: you ask for a job, such as writing a security report or translating a manual, and the configuration decides which engine does it, with which model, how many tokens it can use and how long it can take.
More than 500 AI jobs
From the Instant Help chat to the security, SEO, country, bot or tuning consultants. Each one declared in the gateway.
The heavy work, at night
Reports and long analyses go through nightly tasks and the action queue. Never inside a web request.
The gateway doesn't store the conversation
It receives, distributes and measures. The content of each request doesn't stay in it.
Your data, isolated
Each customer, in their own context
SWPanel's AI works with the data of a single account at a time. Before reading a service it checks that it belongs to whoever asks, and only passes the model what's needed to answer.
Only your portfolio
Each tool checks that the service belongs to your account. The model only receives your portfolio, never another customer's.
A scope filter
It rejects questions about other customers' data and topics unrelated to your infrastructure. In tests it got 80 of 81 messages right.
Only what's necessary
What isn't needed isn't sent: the Leads consultant works without your contacts' email or phone, and passwords are masked in the history.
Data doesn't give orders
Consultants treat the logs and URLs they analyze as data, never as instructions for the model.
How it builds an answer
SWPanel calculates. The AI writes.
The model decides no number. SWPanel measures, cross-references and calculates; the model explains the result in your language and proposes what to do. If what it writes contradicts a figure, a validator catches it and asks again.
Security 360
The consultant starts from the NGFW, the SOC, countries, bots, code analysis, the WAF, actions, resources and service outages.
Server tuning
33 rules calculate the values from live measurements and 7 days of statistics. The model can only cite those rules, not invent others.
See moreInstant Help
Answers questions based on over 1,500 manuals in 7 languages using both semantic and text search simultaneously, ranks the results, and verifies that they truly answer the question.
See moreYou decide
Actions ask for confirmation, proposals await your approval and applied changes can be undone.
How much it uses
Each report, with its AI accounting
AI reports carry an appendix with what it cost to write them. The gateway measures each call and the report converts it into energy, cost and CO2.
What it measures
Model calls, tokens read and written and GPU time.
Energy
The GPU's power times the working time, multiplied by 2.5 to include cooling.
CO2
Total energy times a factor of 250 g of CO2 per kWh.
Cost
The GPU hours plus the electricity consumed.
Doing more with less GPU
Optimized so as not to waste
From 3 minutes to 5 seconds
The answer about manuals went from between 119 and 195 seconds to about 5, reusing what the engine has already calculated.
One model, several roles
The same model classifies the question and solves it: the chat turn dropped from 165 to about 12 seconds.
What's already been answered isn't repeated
Answers about manuals are stored for 7 days and invalidated as soon as the manual changes. Repeated searches don't go back to the GPU.
Only when needed
At most one report per service and night, and only if there isn't one, it's more than 7 days old, something has changed or you ask for it. If you answer “yes” or “no”, it's recalculated without AI.
Faster reports
A full SEO report is written in about 5 minutes; before, it took more than 50.
Translations 44 times faster
The internal translation processes went from about 80 hours to about 2.
What it learns from the whole fleet
What it sees on one service protects the others
Some defenses are decided based on what happens across all SWPanel services at once. Each customer sees only their own; what is shared are patterns and aggregates, never another customer's data.
Global bot lists
A list of allowed and blocked bots for all services, read by the adaptive WAF.
See moreNew bots, detected across the whole fleet
When an unknown bot appears on many services, it's proposed for addition to the catalog and a person reviews it.
Your service, against the average
The firewall report compares your threats with the average of all customers, without showing anyone's.
See moreKnown threats, up to date
Vulnerabilities (CVEs), those already being exploited, IP reputation and attack indicators, in tables that update themselves.
Campaigns before they arrive In preparation
Detecting attacks that sweep through many services at once and bringing defenses forward on those they haven't yet hit.
- 31 AI jobs declared in the gateway
- 459 Manuals in 7 languages
- ~5 s To answer about the manuals
- EU Own Data Centers your data, under GDPR
Frequently asked questions
Questions about the SWPanel AI engine
Your IT, with an AI that never leaves home
Try SWPanel 2027 free and see what its consultants tell you about your own services.