Sonia is trying to make mental health support available through an AI chatbot that users can text or talk to inside an iOS app. The startup presents its product as a way to reach people who cannot, or do not want to, see a human therapist.
The idea is gaining attention because demand for mental health care is high, access is uneven, and therapy can be expensive. But the core question remains open: how far should an AI therapy app go when the science, safety standards, and human role in treatment are still being debated?
What Sonia offers
Sonia describes its product as an "AI therapist" that can discuss a range of mental health topics with users. The app uses generative AI models to analyze what people say during sessions and then respond in the conversation.
The service also applies techniques from cognitive behavioral therapy. After sessions, it can give users "homework" meant to reinforce insights from the conversation, along with visualizations designed to help identify top stressors.
The app costs $20 per month or $200 per year. According to Sonia CEO Chris Aeberli, the product can address issues including depression, stress, anxiety, relationship problems and poor sleep.
Sonia has not received FDA approval. For more serious situations, including people contemplating violence or suicide, Aeberli says the app uses "additional algorithms and models" to detect "emergency situations" and direct users to national hotlines.
The founders behind the AI therapist
Sonia was started by Dustin Klebe, Lukas Wolf and Chris Aeberli. The three met in 2018 while studying computer science at ETH Zürich, then moved to the U.S. together for graduate studies at MIT.
After graduating, they reunited to build a startup around scalable technology. That company became Sonia.
None of the founders have backgrounds in psychology. Aeberli says the company consults with psychologists, recently hired a cognitive psychology graduate and is actively recruiting a full-time clinical psychologist.
Aeberli framed the work as building something closer to a new technology than repackaging an existing care model. Klebe also said Sonia does not view human therapists, or companies that provide mental health care through humans, as competition.
According to Klebe, each response generated by Sonia is supported by about seven additional language model calls in the background. He said those calls analyze the situation from several therapeutic perspectives to adjust, optimize and personalize the therapeutic approach chosen by Sonia.
Why access is the strongest argument
The case for AI therapy apps starts with access. The source article points to a gap in both the ratio of professionals to patients and the cost of treatment compared with what many people can afford.
More than half of the U.S. does not have adequate geographic access to mental care, according to a recent government report cited in the source article. A recent survey cited there also found that 42% of U.S. adults with a mental health condition were unable to receive care because they could not afford it.
Sonia has around 8,000 users and $3.35 million in backing from investors including Y Combinator, Moonfire, Rebel Fund and SBXi. The company is also in talks with unnamed mental health organizations about offering Sonia through online portals.
The app has received quite positive App Store reviews so far. Several users said they found it easier to talk with the chatbot about their issues than with a human therapist.
Aeberli argues that Sonia is easier to begin using than traditional therapy. In the source article, he contrasted the app with finding a therapist, waiting four months, attending at a set time and paying $200.
Where the risks begin
The evidence around chatbot therapy is mixed. One study found that 80% of people who have used OpenAI's ChatGPT for mental health advice consider it a good alternative to regular therapy. A separate report found that chatbots can reduce certain symptoms related to depression and anxiety.
At the same time, the relationship between therapist and client is well-established as one of the best predictors of success in mental health treatment. That human connection is difficult to reduce to a software feature.
The source article also raises practical limits in current chatbot technology. AI systems may give lower-quality advice, and they may miss subtle signals that a human therapist could notice. One example given is an anorexic person asking how to lose weight, while Sonia would not even know the person's weight.
Bias is another concern. Chatbot responses can reflect biases in training data, including Western biases. That can make it harder for a system to recognize cultural and linguistic differences in how mental illness is expressed, especially if English is a person's second language. Sonia only supports English.
There is also a precedent for harm when mental health chatbots are used poorly. Last year, the National Eating Disorders Association came under fire for replacing humans with a chatbot, Tessa, that gave weight-loss tips that were triggering to people with eating disorders.
The transparency test
Sonia sits between two realities. On one side, many people need mental health support and face barriers around geography, cost or willingness to see a human therapist. On the other side, therapy is a high-stakes area where mistakes, missed signals and unclear limits can matter deeply.
Privacy is part of that test. Aeberli says Sonia stores only the "absolute minimum" personal information needed to administer therapy: a user's age and name. But the source article says he did not address where, how or for how long Sonia stores conversation data.
The clearest path for an AI therapy app like Sonia is not to claim that the debate is settled. Its own founders say it is not trying to replace human therapists, and Wolf said the aim is to serve people struggling with mental health who cannot or do not want to access a human therapist.
That makes transparency essential. Users need to understand what the app can address, what it cannot address and when a human professional or emergency resource is the better option. Sonia's future will depend not only on how persuasive its chatbot feels, but on how clearly the company defines the boundaries of what it is building.