A new context
Separate sessions per task or client, with memory settings configured accordingly. Previous conversations are not included unless necessary.
← All services
Safe use of AI
Greater control over your AI and your data.
Protect personal data before a model call. Restrict access and actions. Have key steps checked. AITJE helps to set up the appropriate technical measures.
Discuss your use of AI
From €225Excl. VAT
Safe AI usage begins before you send anything to a model. What information is really necessary? Is this employee authorised to use that data? Can names, contact details and other identifiable details be removed? AITJE helps you incorporate these choices into your workflow.
This is followed by checks on responses, actions and costs. With appropriate guardrails, staff approval, logging and limits, you determine what the solution is permitted to do. ‘AITJE’ also examines where data is processed and stored, and when processing within your own environment is more appropriate.
For whom
- Teams that process customer data using AI
- Knowledge bases with different access rights
- Agents that carry out actions in your systems
- Organisations wishing to gain control over usage and costs
First the data, then the model
Only share what is necessary.
AITJE It can prepare data locally before calling upon an external frontier model. Personal data and other identifiable details are removed or replaced, and the necessary context is selected.
In your own system
NameExample client
Emailklant@example.invalid
QuestionHow do I reset my device?
To the external model
QuestionHow do I reset my device?
Only the necessary question remains.
Removing a name is not always enough. If the remaining text makes someone identifiable, that context also requires an adjustment. Using labels can constitute pseudonymisation; this does not automatically amount to full anonymisation.
Role-based access
Rights determine knowledge.
The application checks who the user is and which documents, categories and text passages they are authorised to use. Only then does RAG retrieve the authorised knowledge.
The model does not therefore see sensitive data straight away. The same access controls apply to linked tools and actions.
Illustrative breakdown of roles
- Internal manualsAvailable for RAG
- Project informationAvailable for RAG
- Financial statementsExcluded from the search
Access is enforced within the application, prior to retrieval and at every tool call.
More than one check
Please also define the rest of the workflow.
Provider & data retention policy
Disable training models where possible and check agreements regarding retention periods or zero-data retention. Sensitive steps may remain local.
Guardrails & manipulation
A filter layer for inputs and sources, restricted tool permissions and control over actions. Limits are also enforced outside the model.
Budget & rate limiting
Budget control prior to calls, limits per user and task, and a maximum number of repetitions. With automatic stops when limits are exceeded.
Check before implementation
Validate output using code, source verification or a second model. For important actions, an employee must first give approval: human-in-the-loop.
Logging & key management
Record user, model, confirmation and action. Agree on log access and retention periods, and protect API keys with restricted permissions.
The measures are tailored to your environment and tested. RAG and an additional assessment model can assist with source and quality control; key outcomes are also subject to human assessment where necessary.
How it works
Step by step, with no surprises.
- 01
Taking stock
Which data goes to which model? Who has access, what actions are possible and how is usage restricted?
- 02
Institutions & locations
AITJE It examines processing, storage, training settings, retention periods and existing controls. Unknown information remains identifiable.
- 03
Selecting measures
A proposal for data minimisation, role-based access rights, guardrails, confirmation and limits. You choose which adjustments are necessary.
- 04
Set-up & testing
Implementation is agreed separately. AITJE tests the chosen measures, including in the event of unauthorised access, unwanted instructions and exceeded limits.
What you get
Tangible results.
- Overview of data flows, providers and data locations
- Insight into access, retention policies and existing controls
- Points for consideration with practical suggestions for improvement
- Following agreed implementation: measures set up and tested
Components
AI-datalocatiecheck
A targeted review of a limited number of known AI services, available provider information and settings. Approximately three hours for a straightforward situation. We reuse findings from a scan; uncertainties remain visible.
From €225
Practical measures
Anonymisation, role-based access rights in RAG, guardrails, employee confirmation, budget limits and logging. The chosen adjustments will be agreed in advance.
Fixed hourly rate
AI-wegwijs
Explanations, guidelines and resources for working with AI smartly and securely, either as part of your collaboration or on its own.
On request
Price
From €225
Excl. VAT
For a simple AI data location check, allow approximately three hours for a limited number of known services. The basic inventory is already included in the AI scan; we reuse previous findings. In-depth technical research and implementation are subject to a separate hourly estimate. The services to be examined and the scope of delivery are agreed in advance.
Discuss your use of AINot included as standard
- A binding legal opinion, certification or guarantee of compliance
- Migration or detailed alternative design
- Extensive follow-up research
Case studies
Does this sound familiar?
From searching through internal knowledge to connecting work processes. Discover the application, the need and the approach.
Estate agency / In-house expertise & website chat
Your firm’s expertise. An assistant that works with it.
For an estate agency, AITJE linked Realworks, email inboxes and Google Drive to its own AI assistant. Staff chat with the knowledge relevant to their role. The website chatbot uses a separate, carefully selected knowledge base from the same environment.
View the approach
Development agency / Assistant, Coder & model management
Ten models. One team. Control and independence.
AITJE set up a shared AI environment for a development agency, featuring Assistent, Coder and ten models to choose from. A shared token pool of €350 per month replaces nine subscriptions to €200. Model usage and token costs per task can be tracked via Asana.
View the approach
Automotive / Voice-controlled workshop
Hey Bob. An extra pair of hands in the workshop.
AITJE, developed a voice-controlled application for a local garage that serves over a hundred customers a day. Six Android tablets connect mechanics to technical information, work orders and their administrative records – simply by asking.
View the approachHow much does an AI data location check cost?
From €225 excluding VAT for a straightforward scenario, estimated at approximately three hours. We check a pre-agreed, limited number of known AI services, settings and available provider information. The basic assessment forms part of the AI scan. We reuse previous findings; additional technical research and measures are quoted separately.
How does ‘AITJE’ help ensure the safe use of AI?
AITJE It maps out data flows, access, data locations and settings, and can implement practical measures. These include deleting personal data prior to an external model call, role-based access rights for the knowledge base, human verification, guardrails, usage limits and logging. The scope of the investigation and implementation is defined in advance.
Can AITJE remove personal data before it is sent to an external model?
Yes. AITJE can incorporate a local preparatory step that removes or replaces names, contact details and other identifiable details, and forwards only the necessary information. Whether the result is truly anonymous also depends on the wider context. Replacing a name with a label does not automatically constitute full anonymisation.
Would it help to start a new, blank chat each time?
A new context for each task or client limits which previous messages are passed on to the next query. AITJE can separate sessions, memory and context. This does not automatically change what the provider itself stores or uses for training: those settings and agreements are checked separately.
Can usage for training or data storage be disabled?
AITJE checks which settings and options are available for the selected service. Disabling the use of data for training is not the same as not storing any data. Zero-data retention may be available for certain APIs and functions, subject to conditions and exceptions. AITJE will investigate what actually applies to your specific use case.
How can you prevent an employee from accessing restricted documents via AI?
The application verifies the identity and role before RAG retrieves data. The search query is restricted to authorised documents, text passages and categories. The same permissions apply to tool calls and other integrations. Restricted content should therefore not first be provided to the model with only the instruction not to share it.
Can AITJE protect against prompt injection and jailbreaks?
AITJE It can combine multiple checks: verifying input and retrieved documents, separating instructions from source material, restricting tool permissions and validating actions. A filter layer or LLM firewall can help with this, but does not stop every attack. That is why access and permitted actions are also enforced and tested outside the model.
Can AI costs and the number of calls be limited?
Yes. AITJE can configure budgets, limits per task and user, rate limiting and a maximum number of attempts. Before a new call, the application can check whether there is still budget available. Provider warnings do not always result in an immediate halt; therefore, the system checks which limits actually cause a block. Ongoing calls and delayed usage reporting are taken into account in the configuration.
When must an employee confirm an action?
This is agreed in advance. For example, before a message is sent, a payment is processed or important data is amended. The employee sees the proposal and the consequences and must give their explicit consent before the action is carried out. The confirmation and execution can be logged.
How are responses and actions monitored?
AITJE It can link responses to authorised sources via RAG and check output against code for format, required fields and permitted actions. A second model, LLM-as-a-Judge, can provide additional oversight. RAG and a second model can help identify errors, but do not rule them out entirely. For important outcomes, human assessment remains an appropriate check.
What is logged internally?
For example, who initiated a task, which sources and model versions were used, what action was proposed, who gave approval and what was carried out. AITJE determines the level of detail, access and retention period. Personal data, full prompts and API keys do not need to be included in a log file by default.
Is running systems locally always sufficient to protect data?
Processing locally can prevent content from being sent to an external model provider. The environment itself still requires appropriate network security, access rights and management. AITJE also checks connections, logging and fallback routes to ensure that confidential data is not sent via an alternative route.
Is a Dutch server automatically secure?
No. A location tells you something, but not everything. ‘AITJE’ also looks at who has access, what is being stored and which settings are enabled.
Does this replace a lawyer or a data protection officer?
No. ‘AITJE’ handles the technical side: where data goes and what measures are possible. For a legal opinion, you’ll need a specialist.
MOVING FORWARD TOGETHER WITH AI
Greater control over your AI and your data.
Protect personal data before a model call. Restrict access and actions. Have key steps checked. AITJE helps to set up the appropriate technical measures.
