Hark launches privacy-first personal assistant that aims to replace the computer

Hark, the startup serial entrepreneur Brett Adcock founded less than a year ago, is releasing Hark Pro today in a broad rollout. The personal AI assistant positions itself as the user interface for models, not the model itself. The service is free with a subscription tier for heavy users, and the core promise is straightforward: the model is trained specifically for computer use, and it operates the machine on your behalf. No talk of AGI or curing cancer — just an attempt to build the operating system for the AI computers of the future.

The product: full-screen, dynamic panels

Hark Pro opens as a full-screen experience with a dashboard-like home screen: a central chat dialog, a feed that suggests proactive actions, and visualizations the company calls "panels" — mini-dashboards for topics you care about. The background adapts to the weather and time at your location. The user connects their digital life — email, calendar, hard drive, credit cards — and the assistant takes on digital tasks alongside classic chat answers.

The demo: what it actually does

Abidur Chowdhury, formerly of Apple and now Hark's head of design, demonstrated the product. The assistant flagged expenses awaiting approval, emails needing a reply, a meeting with this reporter, and offered to book a flight for an upcoming trip. "I know the flight thing has been done to death," he acknowledged with a smile. A more striking moment came when the system detected that a colleague needed to renew a driver's license at the California DMV and completed the process on the agency's website. A notable UX choice: a small window shows the agent navigating the web, a step most assistants perform behind the scenes, in order to build trust that the actions are correct.

Privacy as the differentiator

Hark's pitch rests on a corporate distinction: "We're not here to sell you ads and steal your data," Chowdhury says, alluding to Meta and OpenAI. For these models to be truly useful, the user must hand over their entire digital life, raising real cybersecurity concerns and the question of whether we're comfortable with a company like Meta building detailed profiles of friends and family. Chowdhury counters that the learning is "task-focused" — how Hark takes tasks off your plate, not broad profiling.

The open question: specialized model versus general knowledge

Chowdhury invites critical comparison to competitors like Muse. "Maybe it won't be faster, but the result will be better, and it will work end-to-end without asking you to take over at some point." Hark's model, developed for computer tasks, is faster and cheaper at them, but it's unclear how much general knowledge and frontier LLM capabilities are sacrificed in the process. That is the key question for the entire category: what is the right model for a personal assistant?

Half a promise: a dedicated device in 2027

The software is only half the vision. Chowdhury, who made his name designing devices at Apple, promises an AI-native device to support the software in 2027. For now, they're keeping their cards close to their chest.