JobMentis for coaches: client data, GDPR and AI in Europe
JobMentis runs its servers in Frankfurt and its main AI model in the EU, shares data with US providers only under EU transfer safeguards, lets every client switch off their coach's account access, and erases a deleted account from 158 mapped locations. Each figure here is read from our code.
Read from the code
Three numbers behind this page
158
Storage locations one deletion request covers, with a verification pass afterwards
JobMentis GDPR data map
90
AI instructions that read client text, each carrying a non-discrimination clause that a test enforces
JobMentis prompt registry
48
Test CVs in our last non-discrimination evaluation, across 6 languages
JobMentis bias evaluation, 19 September 2026
Where is my clients' data stored?
The application runs in Frankfurt, the relational database is hosted in Europe and AI text processing runs on Gemini in the EU. The providers established in the United States receive data under the EU-U.S. Data Privacy Framework or Standard Contractual Clauses.
The application
Vercel
Frankfurt, Germany
Accounts, rosters and programmes (PostgreSQL)
Supabase
Hosted in Europe
CVs, stories, notes, files and sign-in
Google Firebase
Google LLC, United States: Data Privacy Framework and Standard Contractual Clauses
AI text processing
Google Gemini on Vertex AI
European Union
AI fallback, if Gemini is unavailable
Anthropic Claude
United States: Data Privacy Framework and Standard Contractual Clauses
Voice practice (mock interviews, the voice CV interview)
ElevenLabs
United States: Data Privacy Framework and Standard Contractual Clauses
Product analytics
PostHog
EU cloud
| What | Provider | Where, and on what terms |
|---|---|---|
| The application | Vercel | Frankfurt, Germany |
| Accounts, rosters and programmes (PostgreSQL) | Supabase | Hosted in Europe |
| CVs, stories, notes, files and sign-in | Google Firebase | Google LLC, United States: Data Privacy Framework and Standard Contractual Clauses |
| AI text processing | Google Gemini on Vertex AI | European Union |
| AI fallback, if Gemini is unavailable | Anthropic Claude | United States: Data Privacy Framework and Standard Contractual Clauses |
| Voice practice (mock interviews, the voice CV interview) | ElevenLabs | United States: Data Privacy Framework and Standard Contractual Clauses |
| Product analytics | PostHog | EU cloud |
Payments, email and the rest of our processors, each with its country and transfer safeguard, are listed in section 5 of our privacy policy.
Who can open a client's account?
The client and the one coach whose roster they are on. While a coach is inside the account the client sees a banner, the visit is recorded on the client's own timeline, and the client can switch the coach's access off at any time.
- Each client sits on one coach's roster. Every coach action that names a client is checked against it, so another coach in your practice, owners included, cannot open your client.
- While you work inside a client's account they see a banner saying so, and the visit lands on their activity timeline with your name, how long you stayed and which features you used.
- The client can switch off your account access at any time under Settings > Coach access. The connection and your past sessions stay; you just cannot go in.
- Your private notes about a client are visible to you only, never to the client.
- A firm that runs JobMentis under its own brand gets its own Firebase project, so its clients' documents are stored apart from everyone else's.
- JobMentis staff can also open an account for support, under the same banner and the same timeline record.
Which AI models read client data, and do they train on it?
Client text goes to Google's Gemini models on Vertex AI in the EU, with Anthropic's Claude as a fallback, and voice practice runs on ElevenLabs. Google and Anthropic do not train on this data under their commercial terms, and no DeepSeek model is used anywhere.
- Every AI call goes through one gateway, so the list on our AI transparency page is complete.
- No DeepSeek model sits in any fallback chain, and a test fails the build if one is added.
- Before any model reads a client's profile summary, it is stripped of contact details, date of birth, age, marital status, children and nationality.
- JobMentis trains no AI model of its own, on client data or anything else.
ElevenLabs does not use voice sessions recorded since 18 September 2026 to train its models: that option is switched off for our account.
How do you keep protected characteristics out of scores?
Each of the 90 AI instructions that read a client's text carries a clause forbidding the model to infer, raise or score on a protected characteristic such as health, religion, union membership, orientation, age or nationality, and a test fails the build if a new instruction lacks it.
On 19 September 2026 we ran the CV score and the job-fit score that ship in the product on 48 test CVs: 8 single-line disclosures in 6 languages, each compared with an otherwise identical control CV. Every variant passed: the scores stayed within the control's own run-to-run spread, and the advice never commented on the disclosure.
- No feature stores a classification of a person.
- Scores are for coaching the person, never for selecting them. JobMentis sends no score, rank or report to an employer. If you also recruit, do not use a client's scores to shortlist them for a hiring client: the product is not built for that use.
What does JobMentis do under the EU AI Act?
JobMentis works for the job seeker: it scores a person's own documents for that person, filters no applications for an employer, tells people when they are talking to an AI, and marks files made with AI help.
- The voice agents (the mock interviewer, the voice CV interviewer, the networking rehearsal) say they are an AI in their first line and whenever asked, in all 6 languages.
- PDF and Word files made with AI help, such as a rewritten CV or a cover letter, carry machine-readable metadata saying so. A CV the client wrote and only edited by hand is not marked.
- No feature recognises emotion from a voice or a face: voice runs at ElevenLabs and we receive text transcripts only.
This section describes what the product does. It is not a legal classification: your own adviser can map these facts to your obligations.
How is a client's data deleted, and what records do you keep?
One deletion request erases a client's account from 158 mapped locations (databases, file storage and connected services such as payments, email and notetakers), then a verification pass checks the stores. We keep a record of processing activities and a data protection impact assessment, both written against the code.
One exception today: voice-practice conversations held by ElevenLabs follow its own retention and are not yet part of that pass.
- Data is encrypted in transit and at rest, and the notetaker and calendar connections you add are stored under their own encryption.
- No form asks a client for health, religion, union membership or similar details. Clients are told how such details are handled if they mention them, at sign-up or on their first visit.
- The notetakers you connect (Grain, Granola, Fireflies.ai, Fathom, SpeakApp, Zoom) and your scheduling tools work for you, under your own agreements with them. A session write-up reaches a client only after you approve it.
Where can I read more?
These pages are public and kept current. Questions from you, a client or a data protection officer go to support@jobmentis.com.
FAQ
Questions coaches ask
Is JobMentis GDPR compliant?
We describe what we do rather than claim a label. JobMentis keeps a record of processing and a data protection impact assessment, uses processors under Article 28 with EU transfer safeguards, runs its servers in Frankfurt and its main AI model in the EU, and erases a deleted account from every mapped store, with a verification pass.
Where is my clients' data stored?
The application runs in Frankfurt, the relational database is hosted in Europe and AI text processing runs on Gemini in the EU. Documents and sign-in sit on Google Firebase, a US company, under the EU-U.S. Data Privacy Framework and Standard Contractual Clauses, as do the AI fallback and voice providers.
Do AI models train on my clients' data?
No. Google and Anthropic do not train on API data under their commercial terms, and ElevenLabs' training option is switched off for our account. JobMentis trains no model of its own, and DeepSeek is used nowhere in the product.
Can a client stop their coach from opening their account?
Yes. Each client has an "Allow account access" switch under Settings > Coach access. Switching it off keeps the connection and past sessions but stops the coach going in. While a coach is inside the account, the client sees a banner and the visit is recorded on their timeline.
Is JobMentis a high-risk AI system under the EU AI Act?
JobMentis is built for the job seeker: it scores a person's own CV for that person, sends no score or report to an employer and filters no applications. Whether a given use is high-risk is a question for your adviser. Using client scores to select people for an employer would change the picture, which is why the product is not built for it.
Want to see it on your own program?
Bring one program and two or three real clients to a call. We build the first schedule with you and show you what the console would flag this week.