DeepSeek has released Harness v0.2 preview with native macOS and Windows desktop apps, the most user-facing step yet for the agent framework the Chinese lab open-sourced on August 13. The installers are live on DeepSeek's official site, and the company is sweetening the launch with 6 yuan of free API credit for users who install the desktop version and sign in before October 6.
The significance is less the version number than the form factor. Until now, running Harness required deploying it through a terminal — a ceiling that kept the tool in the hands of developers comfortable with command-line setup. The v0.2 desktop build is described by the company as out-of-the-box: it comes with preloaded workflows for everyday office and development tasks, a new page for installing and managing plugins, and expanded configuration options for tailoring the agent's behavior.
Early hands-on coverage of the desktop build describes a product that behaves more like a resident assistant than a chat window. Because it lives on the local machine, Harness can run stably in the background, execute scheduled tasks and install new capabilities for itself — the same persistent-agent pattern OpenAI showcased with its Dots and Meta popularized with Muse, but running on the user's own hardware against the user's own files. The app keeps DeepSeek's four agent presets — standard, PTC, minimal and creative — along with support for custom third-party models.
The positioning shift is deliberate. In its official introduction, DeepSeek leads with daily work rather than coding: hand it a Word document to organize, an Excel sheet to analyze and chart, a PDF to extract information from, and a slide deck to assemble at the end. Programming remains part of the pitch, but it is no longer the headline — a notable change for a lab whose models made their name on code and reasoning benchmarks.
The 6-yuan credit promotion is a small number with a loud signal. Chinese observers noted that DeepSeek, long famous for refusing conventional growth marketing, is now running download-and-signup subsidies reminiscent of consumer app launches. The credit is small — well under a dollar — but the tactic marks a change in how the lab is willing to compete for agent users, and it expires October 6, giving the promotion a classic urgency mechanic.
Strategically, the desktop release rounds out a busy stretch in which DeepSeek has been building an agent stack rather than just models: the Harness framework in August, an API registry refresh, and now a consumer-adjacent app that puts the company's name in a dock or taskbar. With OpenAI, Meta and ByteDance all racing to make agents persistent and personal, DeepSeek's countermove is to make its agent local, cheap and portable across operating systems — a distribution bet that costs little and could travel far.
Comments (0)
Log in to join the discussion
Log InNo comments yet