Hark Pro Launches Widely: An AI Assistant That Operates Your Computer
Hark has released Hark Pro widely, free to use with a paid tier for heavy users.
The company detailed the launch on Oct. 6, less than a year after Brett Adcock founded Hark TechCrunch. Hark had previously raised $700 million in Series A funding for what it called a secretive universal AI interface, with the expectation then that its first multimodal models, systems that work with text, images and other input types, would arrive in summer 2026 to power a personal AI platform TechCrunch.
Hark Pro centers on a model trained specifically for computer use, meaning it can click, type and browse on a user's behalf to finish digital tasks instead of only answering in chat. The company says the aim goes beyond an app. Its stated goal is an operating system for future AI computers.
That computer-use work is handled by an agent, software that acts on instructions, called Handoff. According to the company, Handoff runs on its own secure computer with a separate full browser for each user Hark. The work stays visible. Hark Pro shows the agent's web navigation in a small window during a task.
Access is opt-in. If users allow it, Hark Pro can connect to email, calendar, hard drive and credit cards. Chat stays in a single thread with persistent memory, stored context from past talks used to tailor replies, while larger efforts like job searches or trip planning are grouped with Projects Hark.
The interface is a web app, also available on browser, iOS and Android Fast Company. It is full-screen with a central chat box, a feed of suggested actions, and Panels, small visual dashboards. The background shows custom imagery based on local weather and time. The design team is led by former Apple staffer Abidhur Chowdhury.
Two features shape how it works. Action Buttons carry out a chosen task from start to finish, while Panels are user-built mini-apps that can pull in data from services including Strava, Spotify and Venmo Hark. The intent, in the company's words, is building the most personal, human-like intelligence in the world.
Hark is using privacy to stand apart. It says it is not Meta or OpenAI and is not there to sell ads and steal user data. Part of that case is Secured by Hark, which the company says uses end-to-end encryption, a method where only the user can read the data Hark.
Hark Pro joins a market that already includes Muse, Instinct and Dots. Meta describes Muse as a secure, private personal AI agent that proactively helps people meet their goals and suggests ideas Meta. Instinct remains in private access TechCrunch, working through direct texts in iMessage or WhatsApp The Atlantic, an approach that has already raised privacy and security concerns.
The broader context here is familiar to anyone who watched assistants shift from demos to systems that act with permission. Computer-use tools depend on clear permissions, visible activity and ways to recover when something goes wrong. Hark pairs broad access with a visible execution window and a separate browser per user. The test will be steady performance across messy, multi-step tasks like purchases, bookings and file moves, where logins, payments and personal files overlap.
In my view, the operating system claim is the part to watch. We have seen this pattern before, when command lines, launchers and voice assistants each tried to replace the grid of apps with a layer that follows intent. Most struggled to keep developer support and user trust. Hark is starting with trust, through encryption, isolation and an anti-ads stance. If persistent memory, Projects and Panels work well together in daily use, the result would feel less like a chatbot and more like a base for getting things done. The long arc here has given people more control over computers, and a workable delegation model would extend that.
