The Veneer of Authority: Why Onix's 'Expert' AI is a Wellness Gamble

In the current gold rush of generative AI, the most valuable currency isn't just compute—it's credibility. For those of us covering health tech, the red flag is always the same: how is the model being grounded? As BetaKit first reported, Montréal-based startup Onix recently announced a $5 million USD pre-seed convertible note round, framing its approach as a privacy-first revolution. But looking closer at the architecture of their 'personal intelligence' library, it becomes clear that Onix is betting on the *aesthetic* of expertise rather than the evidence of clinical data.

According to reporting from BetaKit, Onix—founded in 2025 by David Bennahum and Nicholas Nadeau—allows users to pay for access to "onixes," which are AI versions of health and wellness practitioners. These range from psychiatrists and family nurse practitioners, such as Ashley Shrader and David Rabin, to holistic nutritionists and former elite athletes. On the surface, this looks like a curated directory of specialists. In reality, it is a content-aggregation play.

As Nicholas Nadeau told BetaKit, the company does not want user data, claiming they have no "data bottleneck" because their experts provide the information. However, the source of this information is the crux of the problem. BetaKit reports that these AI versions are crafted using public appearances, such as podcast recordings, and unpublished works shared by the creators.

Let's be clear: a podcast is not a peer-reviewed clinical trial. A podcast is a medium for storytelling, anecdote, and often, unverified speculation. By training smaller models on the public personas and unpublished writings of "wellness content creators," Onix isn't building a medical knowledge base; it is building a sophisticated mimicry engine. They are essentially wrapping a chatbot in the veneer of medical authority, where the "expertise" is defined by the practitioner's ability to generate content, not necessarily their adherence to evidence-based medicine.

This is particularly concerning given the breadth of the fields involved. BetaKit notes that the experts' fields range from longevity to addiction recovery. In these high-stakes domains, the difference between a "wellness tip" and a clinical intervention is the difference between a lifestyle choice and a medical necessity. While Nadeau told BetaKit that Onix does not provide medical advice and has guardrails to prevent outputs from straying into that territory, the very nature of the product—offering "personalized advice" from AI versions of practitioners—creates a dangerous ambiguity for the consumer.

Furthermore, the incentive structure of the platform encourages the proliferation of "personality" over precision. Onix operates on a revenue-share model where the practitioners set their own subscription fees and receive 70 percent of the revenue, while Onix takes 30 percent. This transforms the practitioner from a healthcare provider into a digital asset manager. When the financial incentive is tied to the number of subscribers attracted to an AI version of oneself, the goal shifts from clinical accuracy to engagement and brand growth.

To its credit, Onix is addressing the massive trust deficit currently plaguing the industry. David Bennahum has acknowledged the volatility of the current climate, noting that AI is perhaps the most disliked new technology polled since atomic weapons. He points to the fact that 83 percent of Canadians express worries about AI and privacy, and that Onix's decision to build under Canadian privacy law is a strategic move to leverage "Brand Canada's" level of trust.

Technologically, the company is making a bold claim regarding privacy. Bennahum told BetaKit that all chat data is stored locally on users' devices with no conversation logs kept, meaning no data makes it back to a data center. Bennahum also asserts that their inference costs are merely "one two hundredths" of what a frontier model requires. If true, this is a significant technical achievement in local AI execution. But privacy is not a substitute for accuracy. A private conversation with an AI that hallucinates wellness advice based on a 2022 podcast episode is still a dangerous conversation.

Onix is positioning itself as the antithesis to Big Tech's data-harvesting model. By avoiding the "vast, constant streams of data" required by frontier labs, they are certainly avoiding the privacy nightmares of the ChatGPT era. But in doing so, they are replacing the "black box" of Big Tech with a "personality box" of wellness influencers.

As the company expands—currently a 10-person team hiring for GTM and sales roles—it must answer a fundamental question: Is an "onix" a tool for health, or is it a digital collectible for the wellness-obsessed? If the library is grounded in the charisma of podcast guests rather than the rigor of clinical data, then Onix isn't innovating health tech; it's just automating the influencer economy. In the world of biotech and health, we don't need more "personal intelligence"—we need evidence.