The idea of recording a session can understandably make both therapists and clients feel uncomfortable. That is exactly why mentalhealthGPT was designed mentalhealthGPT :
The recording is converted to text directly on your device. The audio data is discarded immediately afterward—it is not saved, not transmitted, and never leaves your device.
The only data that is processed is anonymized text—with no names or identifying information.
This text is stored in encrypted form and serves as the basis for the AI—not the recording itself.
ChatGPT and similar tools only know about a case to the extent that you Chat it in Chat current Chat . No memory. No history. No context across sessions.
mentalhealthGPT differently: The transcript of a session becomes a structured entry in the client’s file. With each session, the context grows—and the AI can make meaningful observations about progress, patterns, and next steps.
That’s why the transcript isn’t just an optional feature—it’s the foundation. This page explains how it’s created and protected.
"What was the client's concern in the last session?"
→ No response possible. No memory.
"In Session 3, you used CBT techniques for cognitive restructuring. Session 5 shows a significant reduction in catastrophizing patterns."
→ Context from a real case history.
Every step is carried out in a controlled and transparent manner. Audio, transcripts, and client data are consistently kept separate—only the pseudonym leaves the device, in encrypted form.
The microphone opens in the browser. No app installation is required. The audio data is stored only in your device's RAM and is never written to any storage medium.
The speech is converted to text directly on your device—entirely locally, without an internet connection. No audio leaves your browser. The transcript is generated exclusively on your device.
Before the transcript leaves the device, a locally running AI model automatically identifies all personally identifiable information and replaces it with placeholders: names, diagnoses, medications, and numbers. “Maria Müller suffers from recurrent depression” becomes “[Person-1] suffers from [Diagnosis-1].” What emerges from this step is a text that does not contain any personal references.
Note: The AI receives this pseudonymized text—it does not know any names or identifiable diagnoses. AI responses to clinical topics (e.g., ICD-11 classification) are based on its expertise, not on original client data.
After transcription, the audio data is removed from the device's memory. There is no audio file, no backup copy, and no temporary storage. The recording does not leave the device and no longer exists after this step.
The pseudonymized transcript is transmitted in encrypted form—secured in such a way that no one can read it during transmission. Only this encrypted text is stored: no plain text, no audio, and no identifying information.
The encrypted transcript is assigned to the client's file. It is encrypted with your personal key, which is never stored on our server. Only you can decrypt the content.
For each query, the AI receives the pseudonymized transcript as context: what was discussed, what themes and patterns are discernible, and what interventions were used. The client’s real name and identifying details are not included in this context—the AI understands the case in structural terms, not in a personally identifiable way.
When you read the transcript, unlock your personal key using Face ID or your fingerprint. Pseudonyms are converted to real names locally—only on your screen, without contacting any servers. No aidX employee, no algorithm, and no third party can read your content in plain text.
Unlike ChatGPT and similar tools, each session doesn’t start from scratch. The dossier collects and organizes the patient’s medical history—and provides the AI with the context it needs to make meaningful clinical statements.
Each client receives their own file. All sessions, transcripts, and notes are stored there in an organized manner—encrypted and accessible only to you.
References to previous sessions, identifying patterns over time, consistency in interventions—the AI responds based on the actual course of the case, not on a single prompt.
Each new session adds to the case file. After 3–5 sessions, mentalhealthGPT will be familiar with mentalhealthGPT case history and the interventions used—without you having to re-enter the information each time.
Not every question relates to a specific case. TheChat isChat for case reflection, Supervision, clinical questions, or general considerations—without a transcript, without a case file, and without any reference to clients. Here, the AI operates based on your personal context library and your clinical expertise.
This page is also intended for you as an agent: Here you will find accurate, honest answers to the questions your clients are likely to ask.
No. The audio recording is processed locally in the browser and discarded after transcription. No audio file is saved or transmitted. The voice does not leave the therapist’s device.
Only the pseudonymized transcript is stored—without names or any identifying information. It is end-to-end encrypted. No one except the therapist can read the content, not even aidX AG.
No. The transcription takes place locally in the browser—without an internet connection during recording. The only data that leaves the device is pseudonymized text, encrypted with a key that belongs exclusively to the therapist.
Only the therapist—on a paired device with biometric authentication. Not aidX AG, not third parties, not other system users. The encryption key is stored on the therapist’s device, not on the server.
The stored data is end-to-end encrypted. Without the therapist’s private key—which is stored exclusively on their device—the data cannot be read by anyone. The server does not contain any plaintext data, audio files, or decryptable transcripts.
No. The AI receives only pseudonymized text. Client names and identifying information are replaced locally before the text reaches the AI. The AI is familiar with the case in a structural sense—not with the people behind it.
Recordings always require your clients’ informed consent. This is required by virtually every modern data protection law—whether the DSG (Switzerland), GDPR (EU), or HIPAA (U.S.). When communicated transparently, this consent is not an obstacle but rather the foundation of trust.
Studies show that informed clients generally view mergers positively—especially when it is clear that the practice will remain local and the quality of care will improve.
One practical benefit: If you don't take notes at the same time, you can be more fully engaged in the conversation.
Once clients understand that the recording remains local and that no one other than the therapist has access to it, this discomfort can usually be overcome.
If you don't take notes during the session, you can be more fully present. The transcript is created in the background—the therapist is focused on the conversation.
The client's informed consent is a legal and ethical requirement. mentalhealthGPT does not mentalhealthGPT this conversation—it makes it easier to have.
A suggested phrase for the informational meeting — ready to use, without legal jargon:
"I record our session on my device so that I don't have to take notes and can give you my full attention. The recording stays on my device—it is not transmitted or stored anywhere else. Your name and any identifying information are automatically redacted before anything leaves the device. Only I can access the content, secured by my fingerprint."
Add your personal statement to this sentence, and always obtain the client’s explicit informed consent before the first recording—either in writing or verbally, with the conversation documented.
Please have this sheet available before the first session, or give it to your clients along with the consent form.
Open only this page for printing—without the menu or page content.
14-day free trial — all features, no credit card required. Or try it out in a live demo.