DeepSeek Harness desktop enters global public preview, macOS and Windows installers open-sourced
DeepSeek released a desktop preview of Harness v0.2 on October 2 with macOS and Windows installers, alongside the open-source GitHub repo at around 242,400 stars; positioning shifts from coding-only to everyday work, with new automation task plugins and a plugin management panel, still MIT-licensed.
DeepSeek announced on October 2 that the desktop version of Harness is entering global public preview, with macOS and Windows installers available and Linux access via the npm package @deepseek-ai/dsh; the official download page is live and labelled as a preview. Harness is an open-source agent runtime built on the Cordis plugin architecture, with models, tools, sandbox, and UI all composed of swappable plugins, still MIT-licensed; the GitHub repo has about 242,400 stars and 29,100 forks.
The desktop version shifts positioning from developer tool to everyday work: after signing in or supplying an API key and picking a local workspace, users can browse and process Word, Excel, and PDF files inside the workspace, preview file contents in the chat sidebar, view diffs, or open files in a local editor for editing. The official page's ordering puts "everyday work" before "coding": document organization, spreadsheet analysis, document drafting, presentation building, plus code editing and repository research.
The new release adds an opt-in automation task plugin that lets users turn repetitive work into scheduled jobs that persist even when the desktop is closed; a full plugin management page where users can install, disable, or remove plugins by npm package name; and an experimental creation mode where Harness can write and load new plugins from a chat description. DeepSeek also outlined plans for an official plugin marketplace, sandbox security isolation, Agent Teams for multi-agent collaboration, and personalized long-term memory. Model adapters are plugins too, so users can route Harness to non-DeepSeek model providers.