DeepSeek Harness Officially Launched: The First Agent Product, Fully Customizable Assembly

According to monitoring by 动察 Beating, DeepSeek has officially launched its first Agent product, DeepSeek Harness, with the software package now available on npm. Previously, the public Code Agent evaluation of V4-Flash utilized the minimalist mode built into Harness. It is not just a Coding Agent similar to Codex. Harness supports project management, long tasks, multiple Agents, Skills, online search, and context management, and it also allows users to create different 'Agent Presets.' Tasks such as coding, research, and writing can be configured with different tools, prompts, and workflows. What’s particularly noteworthy is the plugin system. DeepSeek has adopted the design philosophy of 'Everything is a plugin,' meaning that models, tools, prompts, storage, context, and even the interface can be replaced or combined through plugins. Beta users have already developed plugins for long-term memory, interface modifications, and more. Simply put, Codex is more like a pre-built Agent, while DeepSeek Harness is more like a runtime environment for assembling Agents. Users are not just selecting a model; they can also decide what tools the Agent uses, what rules it follows, and how it operates. [Original Link]