DeepSeek AI Releases DeepSeek Harness in Developer Preview: An MIT-Licensed Agent Harness Where Everything is a Plugin

Details

DeepSeek released Harness v0.1 in developer preview and published the full source at deepseek-ai/deepseek-harness under the MIT license. The project ships as the dsh command-line tool. Unlike most agent frameworks that hardcode the agent loop, tool registry, and session store, Harness is built around the Cordis meta-framework with a stated design principle: everything is a plugin. Models, tools, skills, sessions, sandboxes, storage, loops, scheduling, and the UI all sit behind Cordis plugin boundaries. Any component can be selected, swapped, or extended through configuration files without touching Harness source code. This positions it as an agent runtime construction kit rather than a fixed coding assistant. The release ships with four runtime modes: Standard (full coding agent with file editing, shell, search, planning, subagents, and workflows), Code Mode (a TypeScript SDK that lets models compose multi-step operations in one program), Minimal (two tools for benchmarking models in a bare environment), and a fourth configurable mode for custom builds. DeepSeek targets AI-native startups and developer-experience teams inside mid-to-large enterprises. Because it is MIT-licensed and self-hostable, regulated enterprises can run it locally without data-sharing concerns. v0.1 is developer preview — infrastructure in progress, not a production product yet.