Cal AI Co-founder Raises $10M for Persona, an Assistant Built for Messaging
Zach Yadegari, co-founder of calorie-tracking app Cal AI, has raised $10 million for Persona, a new startup building a personal AI assistant.
Vine Ventures led the round, according to TechCrunch. Z Fellows founder Cory Levy and Collective Global took part in the financing. The close was disclosed on Oct. 8.
Background
Yadegari grew Cal AI to more than $30 million in yearly revenue in under two years. MyFitnessPal bought the company in March. Yadegari joined MyFitnessPal in that deal and left in June to start Persona. He spoke on the Builders Stage at TechCrunch Disrupt 2026 about creating viral growth and building on it, a topic tied to Cal AI's fast consumer uptake.
What Persona does
Persona is described as a personal AI assistant in the same group as Instinct and Meta's Muse. It is live as a free beta that works through text messages on iMessage. The company reports a few thousand beta users so far.
The roadmap centers on a wearable band priced at $179 and expected in December. Persona reports five figures in preorder revenue for the hardware. The band is an input and output device for the assistant rather than a standalone computer.
How the band and privacy work
Activation is push-to-talk by design. The band does not listen or record continuously. Contact starts with a button press or a wrist flick. That limit is deliberate. It avoids always-on microphone designs and the permission and power costs that come with them.
Inference, the heavy AI work, runs in the cloud on remote servers. Persona does not run the model on the device and does not store user data on the device. User data is encrypted in transit and at rest, scrambled when sent and when stored, and users can delete their data. The tradeoff is straightforward. Cloud processing simplifies model updates and context management, while putting more weight on secure transfer, retention controls, and deletion enforcement.
Payments and what to watch
Commerce is part of the early design. Persona's agent cannot see users' credit-card details and requires user approval for purchases. Payment is handled through outside providers such as Stripe Link. Persona also plans to show advertised products during shopping research as part of its business model.
Stepping back for builders, Persona is testing three ideas at once. The first is that a consumer agent can spread inside an existing messaging service instead of a new app. The second is that a low-cost wrist trigger is enough to make voice use regular without constant sensing. The third is that users will hand shopping and other tasks to an agent if card data stays tokenized with a third party and every purchase needs explicit approval.
The broader context here is about trust and money. In my view, the tension sits in that third bet. Sponsored listings can fund consumer agents, but they put pressure on ranking logic. Users will ask whether a recommendation reflects fit, advertiser payment, or both. Persona's approval step and keeping raw payment details away from the model help with transaction trust. Ranking transparency will need the same explicit care. If the mix holds, it points to a workable pattern. Keep the endpoint simple, keep inference and storage on servers with clear deletion, keep payments out of the model's view, and monetize assistance rather than access.
