AI Apps for Narcissism and Toxic Relationships: Eight Offers Compared

What kinds of AI support exist for toxic relationships?

Three kinds. Companion chats you talk to like a person: Sophia, Aimee Says, NarcGPT, Toxifree and myProtectify belong here. Analysis tools that evaluate a single message or screenshot and name the manipulation in it: unecho and Nagi work this way, Nagi additionally with a conversation rehearsal in which the AI plays the narcissist. And knowledge apps with tests and a journal, without conversation: NarCure.

The three groups solve different problems. If you want to know what just happened after an argument, a decoder gives you a reading of the message in seconds. If you want to understand why you keep ending up in the same dynamic, you need a conversation that keeps the thread over weeks. If you have to document, for a counselling service for instance, you need a record. Some offers combine two of these.

How do the companion chats differ from each other?

In their foundation. Most offers do not say on their start page which professional basis the AI answers from. Sophia answers from the EMOSOPHIE® by Marion Kohn, who has worked on the root cause of narcissistic patterns for more than 25 years; the foundation can be read as a book. myProtectify names its origin as "by survivors, for survivors". The others state nothing, which does not mean there is nothing, only that it is not stated.

In language. Aimee Says and NarcGPT work in English, Toxifree and myProtectify in German, Sophia in German and English. For a conversation about your own relationship, language is not a detail: nuances such as blame-shifting or the silent treatment are easier to pin down in your mother tongue.

In scope. myProtectify addresses people experiencing domestic violence; that is a distinct, acute situation that also calls for counselling services and the emergency number. Toxifree addresses women after toxic relationships. Aimee Says, NarcGPT and Sophia are broader, Sophia additionally for the question why you keep ending up with the same kind of people.

Where does Sophia take a different approach?

Sophia does not work on the other person’s message, she works with you. She names the patterns when you describe a situation and stays in the conversation over weeks, so you recognise the repetitions. The free tests on this website put into context beforehand what is becoming visible in the relationship. If you want to decode a screenshot instead, unecho or Nagi get you there faster; the two do that better than a conversation.

What Sophia shares with every reputable offer: she makes no diagnoses, does not replace therapy, and is not a crisis service. In an acute situation, your local emergency number and a crisis helpline come before any app.

How do I recognise a reputable AI offer?

By five things you can check on the start page or in the store listing within minutes. First: the offer names its professional foundation, not just a promise. Second: it says clearly that it is not therapy and not a diagnosis, and it names an emergency number. Third: it explains what happens to your conversations, who processes them, and whether they are used for advertising. Fourth: the price is stated, including subscription terms. Fifth: it promises no cure and no guarantee that the other person will change.

A sixth point is less visible but decisive: how does the offer treat your perception? A good tool helps you trust your perception again. A bad one replaces the narcissist’s voice with the voice of the app.

How was this comparison made?

On 8 September 2026 we asked ChatGPT and Perplexity which AI apps they recommend to people in toxic relationships and included the offers they named: Aimee Says, NarcGPT, Nagi, Toxifree and myProtectify. Added to these are unecho and NarCure, which appear before or next to our own app in the German App Store when searching for "Narzissmus". All details come from the providers’ start pages and the store listings of the same day.

Sophia is our own product, and we say so openly. That is why we only compared characteristics that can be verified and awarded no ratings. If a detail is no longer correct, let us know and we will fix it.