A clear purpose
Identify the business task and the minimum information needed to complete it. Use synthetic or redacted examples when exploring a workflow, and avoid unnecessary personal data in prompts and logs.
Privacy & security
Your data deserves care. From the first design.
Useful AI starts with a clear understanding of your information, your people and your responsibilities. We work with your team to build privacy, security and human oversight into the way your software works.
Discuss your requirementsPrivacy by design
Start with the data you actually need.
Before connecting a workflow, we work through its purpose, the information it needs and who should be able to use it. These decisions shape the design, the default settings and the review before launch.
Official sourcesEDPB: privacy by design & by default
Access with a reason
Agree access for each role, including administrators and support. Keep private information behind permissions and decide explicitly which sources an assistant may use.
A defined lifecycle
Set retention and deletion requirements for records, conversations, logs and backups. Include how people can exercise their rights and who is responsible for handling requests.
Security by design
Practical safeguards. Clear boundaries.
Our application provides a foundation of permissions, private records and constrained AI assistance. We review the controls needed for each customer environment with the people responsible for operating it.
Protected access
Private operational records and documents require authentication and role permissions. The application limits repeated failed sign-ins. Stronger identity requirements, including SSO and MFA, are scoped for the deployment.
Separate environments
The deployment tooling gives each customer instance its own database, storage and application secrets. Hosting arrangements and infrastructure access still need their own review.
People approve the result
The CMS assistant prepares suggestions from published website content and the signed-in user’s conversation. It has no tools to publish pages, execute commands or read private operational records. A person reviews and publishes through the CMS.
Encryption that fits the data path
The public website uses HTTPS. For customer workflows, we agree transport protection, storage and backup encryption, and key management across the actual services involved.
Controlled changes
Application changes go through automated checks before deployment. Changes to tracked operational records are recorded in an activity log. Logging coverage and retention are agreed for the workflow.
Maintenance & recovery
Define who handles updates, incidents and recovery. Backup schedules, restore testing, monitoring and response commitments are part of the operating scope agreed with your team.
GDPR / DSGVO
Data protection is part of the project.
GDPR and DSGVO name the same EU data protection regulation. Where it applies, the work includes lawful processing, transparency, data minimisation and appropriate security. We also consider the Swiss Federal Act on Data Protection (FADP / DSG) where applicable.
Roles & agreements
Clarify who determines the purposes and means of processing and who acts on instructions. Where Mandem acts as a processor, the project requires a data processing agreement, documented instructions and agreed subprocessors.
Rights & accountability
Define the lawful basis, information notices, retention and the process for access, correction, deletion and other applicable rights. Assess whether a data protection impact assessment is required before processing begins.
Transfers & providers
Review processing locations and international transfers, including support access. Use an applicable adequacy decision or appropriate safeguards where required, with any necessary transfer assessment.
EU AI Act
Assess the use. Keep people in control.
The EU AI Act takes a risk-based approach. Obligations depend on the system’s intended use, the organisation’s role and the applicable rules. We work through these questions for the proposed workflow alongside data protection requirements.
Classify before launch
Identify provider and deployer responsibilities, screen for prohibited practices and assess whether the use is high-risk. Sensitive applications such as recruitment or access to essential services need particular scrutiny.
Make AI understandable
Plan the disclosures and content labelling required for the use case. Help the people using and supervising the system understand its limits, check outputs and recognise when to escalate.
Match oversight to the risk
Define human review, evaluation and documentation for the workflow. Check the current application dates and transitional provisions before launch and when the system’s purpose changes.
Hosting & model choices
Know where your information goes.
Where an application runs and where an AI model processes a prompt are separate decisions. We map the full route with you, including connected services, logs, backups and support access.
Local or hosted
A local application can still send information to an external model. A fully local workflow needs compatible local processing and a review of every supporting service and connection.
Provider terms matter
For a proposed model service, review its processing locations, retention, training use, subprocessors and contractual terms before approving the information it may receive.
Training use is an explicit decision
Include whether prompts, files or outputs may be used to improve a model in the provider review. Any no-training or retention commitment must be supported by the selected service’s contract and configuration.
Working with your team
Agree the details. Verify the setup.
For your project, we agree the data flows, permitted access, providers, review steps and operational responsibilities before connecting customer information. The exact safeguards and commitments are documented for that deployment; this page describes our approach and does not establish a certification or a blanket compliance guarantee.
Before launch
Review the processing agreement, provider list, risk assessment, permissions and recovery requirements with your IT and data protection contacts. Record what has been verified and what still needs attention.
As the workflow changes
New data sources, model providers or automated actions can change the risk. Include a review of access, processing arrangements and responsibilities when expanding the system.
An open conversation
Let’s work through
your requirements.
Bring your IT, security or data protection questions. We’ll work through the proposed setup and the responsibilities with your team.
Discuss your requirementsPrivacy questions or a security concern?
hello@mandem.ai ↗Website privacy notice →Hosting & AI options →Regulatory sources reviewed 6 September 2026.