mentalhealthGPT
A Swiss clinical assistant built for licensed mental health professionals. It transcribes sessions locally in the browser, replaces client identifiers with placeholders before any AI processing, and keeps an encrypted case file that informs every later answer.
What is mentalhealthGPT?
mentalhealthGPT, presented on its own site as my private mentalhealthGPT, is a clinical work environment for mental health professionals, published by the Swiss company aidX AG in Pfäffikon ZH and founded by Dr Peter Wildhaber, who combines psychology training with AI research at ETH Zurich. The terms restrict it to licensed therapists, psychiatrists, psychologists and equivalent roles, or to trainees working under supervision.
The centre of the product is the Dossier: one encrypted workspace per client holding sessions, transcripts, notes, analyses and reports. The AI answers against that accumulated history rather than against an isolated prompt, which is the difference the vendor draws with general chatbots — it names ChatGPT and DeepSeek directly. Around the Dossier sit a Context Library organised in four layers (a psychiatric knowledge base covering ICD-10 and ICD-11 criteria, therapy guidelines such as CBT, DBT, ACT and EMDR, pharmacotherapy and crisis standards; then institutional, personal and case context), a free reflection chat, and a Supervision space still under development.
The privacy design is the other half of the pitch. Session audio is transcribed by a model running in the browser. A local named-entity step then swaps names, places, diagnoses, medications, social security numbers, phone numbers, IBANs and birth dates for neutral placeholders before any text leaves the device. Content is encrypted client-side under a three-tier key model, and the mapping back to real names is unlocked biometrically on the user's own device. Sign-in uses the smartphone as a passkey rather than a password, with QR pairing for additional devices.
Practically, the application ships 28 clinical document templates, from SOAP and DAP notes and a mental status exam to a suicidality risk assessment, a safety plan, a crisis intervention note, and country forms for Switzerland, Germany and the United States. Hosting and clinical transcription run on Infomaniak in Switzerland. The interface and the answers are available in English, German, French and Spanish. aidX states plainly that the tool is not a medical device under the EU MDR: it is decision support, and the clinical decision stays with the professional.
What it does
- Transcribe a recorded or uploaded therapy session locally in the browser, with speaker separation for up to four people
- Replace names, places, diagnoses, medications, social security numbers, phone numbers, IBANs and dates of birth with placeholders before anything reaches the AI
- Keep every session, note, document and analysis of one client in a single encrypted case file the AI reasons over
- Analyse a session in five styles: short version, structured summary, supervision question, evidence-based intervention ideas, extracted themes
- Draft clinical reports from 28 templates, including suicidality risk assessment, safety plan and crisis intervention note
- Feed a personal or institutional context library so answers follow your own methods, guidelines and wording
- Suggest ICD-10, ICD-11 and DSM-5 diagnostic hypotheses from the documented case history
When to use mentalhealthGPT / When not to
A quick filter to help you decide if mentalhealthGPT is the right fit.
When to use mentalhealthGPT
- Psychotherapists, psychiatrists and clinical psychologists in solo practice who lose evening hours to session notes
- Small group practices that need consistent documentation without handing client material to a general-purpose chatbot
- Psychiatric clinics, therapy centres and university services needing team accounts, an institutional knowledge base and an audit trail
- Clinicians in Switzerland and Germany who must produce insurer and pension-fund reports (IV, KVG, SUVA, DRV, GKV, PTV 3 and 3a)
- Practitioners bound by professional secrecy who want AI support without transmitting identifiable patient data
When not to use mentalhealthGPT
- People seeking help for themselves: this is a professional workbench, not a therapy chatbot or a crisis service
- Anyone in acute distress, since the site publishes no emergency numbers and no crisis escalation procedure
- Unlicensed practitioners, as the terms restrict the service to licensed professionals and supervised trainees
- Teams wanting a mobile app or an API, because the product runs in the browser only and publishes no developer documentation
- Clinicians who cannot record audio properly, since usable transcripts depend on a wireless worn microphone
How to use mentalhealthGPT
A typical end-to-end flow, from setup to results.
- Start the 30-day trial from the sign-up link; no credit card is required and nothing is charged when it ends
- Pair your smartphone by QR code so it becomes your security key through a passkey unlocked by Face ID or fingerprint
- Create one Dossier per client; both the label and the contents are encrypted end to end
- Record the session in the browser or upload an audio file in MP3, M4A, WAV or WebM, up to 500 MB
- Review and correct the transcript before it is processed; the audio is dropped from memory once transcribed
- Run the session analysis in whichever of the five styles you need; the result is ephemeral and not stored
- Upload your therapeutic frameworks, guidelines and templates into the Context Library so answers match your practice
- Generate reports from the Dossier and check the pseudonymisation before reusing any passage externally
- Connect Zoom if you work remotely, so meetings you host are captured and transcribed through the same pipeline
- Manage or cancel the subscription under Settings, Subscription, Manage subscription, which opens the Stripe portal
Pros & Cons
Pros
- A privacy architecture described in unusual technical detail: local pseudonymisation, client-side encryption, and keys that never leave the device in clear
- Swiss hosting and transcription with Infomaniak, explicitly framed against the nDSG and professional secrecy under article 321 of the Swiss Penal Code
- Longitudinal memory per case, so the assistant reasons over the whole documented history rather than a single prompt
- Twenty-eight clinical templates including real administrative forms for Switzerland, Germany and the United States
- Clinical guardrails written into the templates themselves: supervisor review is flagged as required on suicidality risk assessment and crisis intervention notes
- An unambiguous regulatory statement — not a medical device, the clinical decision stays with the professional — plus a printable consent sheet for the client conversation
- A 30-day trial with no credit card and no automatic charge at the end, and no tracking cookies, marketing analytics or behavioural profiling
Cons
- No Article 27 EU representative, no named data protection officer and no data processing agreement offered to the practitioner, who is the controller of the patient data
- The site contradicts itself on infrastructure: the security page rules out AWS, Azure and GCP while the privacy policy lists Google Cloud Vertex AI as an essential sub-processor
- It also contradicts itself on training: the security page says user data is not used for model training, while a support answer sells institutions a model trained on their documentation
- No crisis protocol and no emergency numbers are published, although the marketing claims crisis screening and BERT-based detection of crisis indication
- ISO 27001-aligned and HIPAA-ready describe alignment only; no certification, external audit or penetration test report is published for a product whose whole promise is cryptographic
- The commercial offer is still young: only the limited Founding Member plan is open, while Supervision and the Professional plan are announced but unavailable
- No minimum age is published, no content retention period is stated beyond a subscription tier history limit, and there is no mobile app, no public API and no social presence
Pricing & Plans
There is no permanent free plan. Access begins with a 30-day free trial covering all available features, with no credit card required and no automatic charge when it ends. The lowest paid entry point currently open is the Founding Member plan at CHF 111.00 per month, or CHF 1,111.00 per year, billed alternatively at EUR 122.00 or USD 144.00 monthly depending on location. The Professional plan is announced at CHF 189.00 per month but is not yet on sale. Institutional pricing is quoted per seat on request. Amounts were recorded on 31 August 2026 and the displayed currency follows the visitor's location.
- Free Trial — 30 days
- all available features
- no credit card
- up to 3 Dossiers
- fair use
- ends automatically with no charge
- limited places
- full access to current and future features
- unlimited Dossiers
- clinical reports in PDF and Word
- multi-device
- early access
- direct influence on the roadmap
- rate locked permanently
- unlimited Dossiers
- Supervision Mode
- clinical reports
- up to three devices
- email support
- everything in Professional plus institutional Context Library
- team accounts
- admin dashboard
- audit trail and compliance reports
- GDPR
- nDSG and HIPAA documentation
- dedicated onboarding
- priority support and an account manager
Data, GDPR & hosting
A consolidated view of how mentalhealthGPT handles your data.
GDPR overview
The commitments are concrete but incomplete. The privacy policy carries a section headed Your Rights (GDPR / nDSG) listing access, rectification, erasure, portability, objection and the right to lodge a complaint with a supervisory authority, with privacy@aidx.ch as the contact. aidX AG is named as controller at its Pfäffikon address. The security page places the design under the Swiss nDSG, the GDPR and Swiss Penal Code article 321 on professional secrecy, and describes ISO 27001-aligned practices. Sub-processors are published. What is missing matters for an editor outside the EU: no Article 27 EU representative is designated, no data protection officer is named, and no data processing agreement is offered to the practitioner, who is the controller of the patient data. No minimum age is published anywhere. ISO 27001-aligned and HIPAA-ready describe alignment, not certification.
Who owns the data?
The security page states that users and their organisations keep full ownership of their data, and that encrypted content stays under the user's control. aidX AG says it does not access clinical content, does not analyse session content, and does not use user data for model training: content is encrypted in the browser and the keys are derived on the device, so the servers hold ciphertext only. Ownership stops at the content, however. The terms state that all content, models and software belong to aidX AG or its licensors, and that no rights pass to users. Responsibility runs the other way: the practitioner, not the vendor, must ensure any patient data processed complies with professional secrecy and data protection law.
Reuse rights
Clinicians may reuse what they produce without asking permission. Case files, transcripts, notes and generated reports belong to the user, can be exported as PDF or Word, and can be deleted manually at any time; after cancellation the encrypted content stays reachable until the paid period ends, and aidX says it cannot open it afterwards, so exporting first is advised. What users may not do is reverse-engineer or copy the service, share credentials, or process patient data without appropriate consent. On the vendor's side, the privacy policy of July 2026 lists account email, anonymised usage logs, encrypted content and technical data, used to run the service, handle Stripe billing, send transactional email and prevent abuse. Two statements sit uneasily together and deserve a direct question to the vendor: the security page rules out AWS, Azure and GCP while the privacy policy names Google Cloud Vertex AI in the EU region as an essential sub-processor, and the same security page rules out training on user data while a support answer offers institutions a private model trained using your documentation.
Data retention & training
Hosting summary
Core infrastructure, databases and stored transcripts are hosted by Infomaniak Network SA in Switzerland, which also performs clinical transcription; the resolved server address belongs to Infomaniak in Geneva. Swiss jurisdiction and the nDSG are claimed explicitly. The privacy policy adds that some clinical text analysis is processed transiently on Google Cloud Vertex AI in an EU data region, and states that no clinical data is stored outside Switzerland; decryption keys are described as held in Switzerland and separated from the processing infrastructure. Four sub-processors are named: Infomaniak in Switzerland, Google Cloud Vertex AI in the EU region, Zoom for the optional meeting integration, and Stripe for payments. Zoom audio is relayed in memory only and never stored raw. One caveat: the security page claims no US hyperscale cloud provider is used and names AWS, Azure and GCP, which the privacy policy's own sub-processor list contradicts. The privacy policy, as the more specific legal document, is what is reported here.
Things to keep in mind
Risks and trade-offs to weigh before adopting mentalhealthGPT.
- No crisis protocol, emergency number or escalation path is published, while the marketing claims crisis screening and automated detection of crisis indication — a gap between what is promised and what is documented
- Clinical oversight is asserted but not evidenced: the vendor cites psychiatrists, psychologists and clinical ethicists and a human-in-the-loop review, yet names no committee, method or frequency, and the About page identifies only the founder
- The tool is not a medical device and has passed no conformity assessment, so a clinician who leans on its diagnostic suggestions or intervention ideas carries the full responsibility alone
- Automation of clinical reasoning erodes it: drafting notes, hypotheses and reports by prompt over months can dull the observation and formulation skills that make the record worth keeping
- Informed consent stays with the practitioner; the vendor supplies a printable sheet but does not enforce its use, and recording a session without proper consent is the practitioner's breach, not the vendor's
- Zero-knowledge cuts both ways: because aidX cannot recover the keys, losing device access without completing the pairing recovery flow risks losing the case history, and after cancellation nobody can reopen the content
- No minimum age is published although the tool processes records of clients who may be minors, and no content retention period is stated beyond an unspecified subscription tier history limit
Setup & Integrations
Technical difficulty
Low, with one unavoidable step. Nothing is installed: the product runs in a modern browser on a computer, laptop or tablet. First access requires pairing your smartphone by QR code so it becomes your passkey, unlocked by Face ID or fingerprint; further devices use the same flow, and any device can be revoked without data loss. The real prerequisite is hardware — the vendor states that a wireless worn microphone is needed for accurate transcription and speaker separation, and that the AI cannot reconstruct what the microphone missed. Dedicated onboarding comes only with the institutional plan.
Deployment
Integrations
Supported languages
Behind mentalhealthGPT
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement mentalhealthGPT.
Frequently asked questions
Who is mentalhealthGPT for?
Is it a medical device?
Can the vendor read my patients' data?
Where is the data hosted?
Is my data used to train AI models?
What does it cost and is there a free trial?
Is there a mobile app or a public API?
What happens to session audio, and how long is data kept?
Does it handle crisis or suicide risk?
Should you pick mentalhealthGPT?
mentalhealthGPT is one of the more seriously built tools in a category where seriousness is not the norm. It is narrow on purpose: a case-centred workbench for licensed clinicians, not a wellbeing chatbot, with a privacy design that goes further than a policy page — local transcription, local pseudonymisation of identifiers, client-side encryption, keys that stay on the device, Swiss hosting. The twenty-eight clinical templates are the tell that practitioners were involved: insurer and pension-fund forms for Switzerland and Germany are not something a generic AI startup ships. Two details deserve credit. The regulatory statement is unambiguous — not a medical device, the decision stays with the professional — and suicidality risk assessments and crisis intervention notes carry a built-in flag that supervisor review is required.
The reservations are about what is not written down. For a Swiss company inviting EU practitioners to process patient data, the absence of an Article 27 representative, of a named data protection officer and of any data processing agreement is a real gap, since the therapist is the controller. No minimum age appears anywhere. The site contradicts itself twice on points that matter: it rules out Google Cloud on one page and names Vertex AI as an essential sub-processor on another, and it rules out training on user data while offering clinics a model trained on their documentation. Nothing about the cryptography has been externally audited, and ISO 27001-aligned is not a certificate.
The offer is also early: only the limited Founding Member plan is actually on sale. Worth a trial for clinicians who document heavily, with the contradictions above put to the vendor in writing first.
- Choosing a selection results in a full page refresh.
- Opens in a new window.