Shadow AI
Contracts, customer data and figures get pasted into consumer chatbots. No logging, no data processing agreement, no idea where any of it ends up.
Your own AI, in your own server room. Independent of outside parties.
werqly places an AI node in your own server room. The open-weight models run there and nowhere else: no AI vendor in the processing, no data leaving your building. Every AI automation runs on it, and your IT department decides in werqly AI Control who may do what, what gets logged and where the limits are.
- Runs in your own server room
- Open-weight models only
- Independent of outside parties
- GDPR & EU AI Act
Running on your own node
Role: Finance
- Llama 70B
- Mistral Large
- Qwen 32B
Safeguards active
Insight: anonymous (default)
- National ID detection
- Source code filter
werqly Node — your own AI node, in your own server room
What is going wrong in your organisation?
AI tools without an owner
Every department buys its own AI subscription. Data in eight jurisdictions, eight vendor assessments, and nobody who sees the total — or can steer it.
No learning curve
The organisation uses AI but learns nothing from it. Which questions keep coming back? Where do safeguards fire? Which process is begging for automation? Nobody knows.
BYOD worked for phones. It does not work for AI: what flows through it is your most valuable asset — and what you learn from it, your edge.
Sectors
In which sectors is confidentiality a legal duty?
Law firms, healthcare providers, financial organisations and public bodies work with data that does not belong in a public AI tool. For each sector we show what the rules look for, how werqly fits, and where it does not.
Law firms and notaries
It is not about AI, it is about professional secrecy. Case files and client data never leave your firm.
View the sector →Healthcare
Patient data, national hosting rules and medical confidentiality: AI that runs in your own server room.
View the sector →Financial sector
Client data, DORA and the AI Act: AI whose location you can evidence.
View the sector →Local government
Residents’ data, algorithm registers and the AI Act: AI you manage yourself.
View the sector →How it works
How does it work?
- 01
Your node goes live
werqly installs the node in your server room and connects it to your identity provider and werqly AI Control. From day one the models run on that node, and nowhere else. No server room? Then the node sits in a locked rack in our European datacentres.
- 02
IT sets the boundaries
In AI Control your IT department decides per role which models are available, which safeguards apply, how much insight the organisation gets into usage, and which automations who may start.
- 03
Automations go live
Which processes come first follows from the TIMA Scan. We build the automations — or a partner of your choice does — and host them on your node. Employees use them through the browser.
werqly AI Control
Everything your admin needs. Nothing a department can route around.
One console for the whole organisation. What is not in there does not run on your node.
Roles and permissions
Groups from your identity provider, wired straight to models, automations and safeguards.
Models per role
Finance gets a different model than the service desk. You set which model, which context length and what may leave.
Safeguards and DLP
Block patterns such as national IDs, customer numbers, IBANs or source code before they reach a model. Configurable per role.
Insight into usage
Three modes: anonymous, per department, per user. Anonymous by default. See below how that works.
Learning from questions
Recurring questions and patterns become visible — aggregated. That is your shortlist for the next automation.
Managing automations
Switch on and off, versions, who may start what, which data an automation may touch.
Capacity per department
Divide your node’s compute. Batch work outside office hours, priority for what really matters.
Audit log and SIEM
Every prompt, model and automation run, timestamped. Exportable to your SIEM.
Learn from AI usage without turning the works council against you.
How much your organisation sees of what employees ask AI is a policy decision — not something a vendor decides for you. In AI Control you choose the mode. The default is anonymous.
Anonymous
Trends per department and per automation, recurring themes, where safeguards fire. Never traceable to a person.
Per department
The same insights, plus usage and patterns per team. For steering capacity and deciding which automation comes next.
Per user
Individual prompts visible, only after approval by two admins and with a record in the audit log. For security incidents, not for daily practice.
Every mode is recorded in policy, visible to the DPO and presentable to the works council. Switching modes is itself logged in the audit trail.
Built for the question your auditor asks.
werqly exists because AI belongs in a managed environment. Not as a tool with a checkbox, but as infrastructure that is yours alone and that you can evidence.
Models run on your own node
Prompts, documents and the output of automations are processed on a node in your own server room that belongs to your organisation alone. No outside party in the processing. If you choose werqly Cloud, that node sits in a locked rack in our European datacentres: never shared, never at a hyperscaler.
Open-weight models only
Llama, Mistral and Qwen, hosted locally. No step in the processing reaches an external model vendor — not even as a fallback.
No training on your data
Your prompts and documents are never used to train models. Not by us, and not by a third party.
Remote management, visible and severable
werqly manages the node over an encrypted management link to AI Control. It carries policy, updates and the insights you have chosen to release. When the node sits in your server room, you can see that link and cut it.
Encrypted, keys with you
Encryption on storage and on every connection. Key management in your environment, with rotation.
Complete audit trail
Who, what, which model, which automation, when. You set the retention period.
Automations
Built for your organisation. Hosted on your node.
werqly does not ship a catalogue of standard integrations. Which processes your organisation should automate follows from the TIMA Scan. We build those automations — or a partner of your choice does — and they run on your node, never at a third party.
For example
Financial analysis
Review monthly figures, flag deviations, prepare commentary.
Invoice and document processing
Read, classify and stage incoming documents for approval.
Contract and policy review
Mark deviations from your standard terms before the lawyer looks at it.
Reports from ERP and CRM
Assemble periodic reports from your own systems, in your own format.
Classifying customer questions
Route incoming questions and stage a draft answer for the employee.
Internal knowledge base
Make policies, procedures and handbooks searchable, with sources.
Calculate your private AI costs.
Enter your work email and number of employees. We work out four setups for you straight away: in your own server room or in our datacentre, self-managed or with werqly Pro management. werqly is not a cheap AI subscription: you pay for your own hardware and full control.
- 1Enter your email and number of employees
- 2Compare four proposals
- 3Request a quote straight away
Working out the proposals…
Your own server room
Your node in your own server room or rack
werqly Cloud
No server room? A locked rack in our EU datacentre
werqly Pro management
Your own server room
We operate the node and the models
— per month
—
- Data never leaves your premises
- No outside party in the processing
- Updates, monitoring and support by werqly
werqly Pro management
werqly Cloud
We operate the node and the models
— per month
—
- For organisations without a server room
- Dedicated node in a locked rack, never shared
- Updates, monitoring and support by werqly
Self-managed
Your own server room
Your IT department operates it; we supply software and updates
— per month
—
- Data never leaves your premises
- No outside party in processing or management
- Full control with your IT department
Self-managed
werqly Cloud
Your IT department operates it; we supply software and updates
— per month
—
- For organisations without a server room
- Dedicated node in a locked rack, never shared
- Your IT department runs policies and automations
Indicative amounts, excluding VAT. Custom automations are quoted separately. Only the quote is binding.
werqly next to a hyperscaler AI suite.
| werqly | Typical cloud AI suite | |
|---|---|---|
| Where does the model run? | In your own server room, or a locked rack in our EU datacentre | The vendor’s datacentres |
| Which models? | Open-weight: Llama, Mistral, Qwen | The vendor’s closed models |
| Jurisdiction | EU: your premises or our own datacentre | Often a US parent company |
| Training on your data | Never | Contractually excluded, technically unverifiable |
| Dependence | No outside AI vendor in the processing | Dependent on the model vendor and its terms |
| Automations | Custom, run on your node | Standard, run at the vendor |
| Insight into usage | Three modes, you choose | Limited, varies per vendor |
| Management | Your IT department, one console | Per tool, different per department |
From the knowledge base
Background for CISOs, IT managers and compliance officers.
What is private AI? Meaning, forms and when you need it
Private AI means language models run on infrastructure that belongs to your organisation alone. What it is, which forms exist and when it is the right choice.
Read the article →Is ChatGPT GDPR-compliant for your organisation?
Can your organisation use ChatGPT under the GDPR? The difference between free and business use, when it becomes a data breach and what you need to arrange.
Read the article →Shadow AI: what to do when staff use ChatGPT on their own
Shadow AI is AI use outside IT’s view. Why a ban alone does not work, what happened at the Dutch city of Eindhoven, and how to bring it under control in five steps.
Read the article →What security teams ask first.
Do the models really run on our own node, or somewhere behind a European endpoint?
Really on your node. The open-weight models sit on the GPUs of a node that belongs to your organisation alone — in your server room, or with werqly Cloud in a locked rack in our European datacentres — and process prompts, documents and automations there. No call goes out to an external model vendor, not even as a fallback.
What exactly travels over the link to AI Control?
Policy and configuration towards the node; software and model updates; and back only the insights you have released in the insight policy, plus technical health data of the node. No prompts, no documents.
Are open-weight models good enough for our work?
For the vast majority of office work and process automation — summarising, classifying, analysing your own documents and figures — the gap with closed models is small and access to your own data matters more. For work where that is not true, werqly is not the right answer. We would rather say so up front.
What does our server room need?
One rack position of a few rack units, a standard power connection and a network link to your office network. You get the exact specification with the quote. Without a server room, the node can sit in a locked rack in our European datacentres.
Who builds the automations?
werqly, or an implementation partner of your choice. What needs building usually follows from the TIMA Scan. The automations run on your node and are managed in AI Control, regardless of who built them.
What does the admin see of what employees ask?
You decide that in the insight policy. The default is anonymous: trends and patterns, never traceable to a person. Per department and per user are optional modes; the latter only with four-eyes approval and a record in the audit log.
Can we start small?
Yes. Many organisations start with one automation for one department on their own node, and expand from there. The calculator shows which setup fits your size.
Why does werqly cost more than an AI subscription?
Because you are not renting a shared service but running your own hardware with your own models on your own premises. You pay for control: no outside AI vendor in the processing, no data leaving your organisation, and management you can show to an auditor. For organisations where security and compliance weigh more than the lowest price, that is the trade-off.
Let’s see whether werqly fits your organisation.
In a 45-minute call we walk through your situation: which AI is already in use, how the node fits your server room, and which processes qualify first. No obligation.
- A proposed date within two working days
- Security datasheet sent ahead of the call
- Intake calls are running now