How to Install DeepSeek Harness in Minutes and Run Your Own AI Agent

Want a powerful, open-source AI coding agent you can run locally? DeepSeek Harness (dsh) makes it straightforward.

DeepSeek Harness is an MIT-licensed agent runtime from DeepSeek that turns a language model into a real agent. It can read and edit files, run shell commands, search the web, manage sessions, and keep working on multi-step tasks inside a project folder. The key design principle is “everything is a plugin”—models, tools, the agent loop, sandboxes, and even the UI can be swapped or extended.

The fastest way to get started is one command (Node.js required):

npx @deepseek-ai/dsh web

This launches a local web interface (usually at http://127.0.0.1:3080). Open it in your browser, add a DeepSeek API key under Settings → Models, choose a workspace folder, and you’re ready to give the agent tasks.

Prefer a desktop app? Official installers are available for Windows and macOS. Developers can also clone the GitHub repo and build from source with pnpm for the latest changes or plugin development.

Once running, you can use Standard mode for full capabilities, Minimal mode for benchmarking, or Creator mode to let the agent help build new plugins. It also supports OpenAI-compatible endpoints if you want to point it at other models.