DeepSeek Harness: Everything Is a Plugin
DeepSeek Harness (dsh) is an open-source agent harness developed by DeepSeek AI. The model thinks; the harness acts. In this framework, models, tools, skills, sessions, sandboxes, storage, agent loops, schedulers — even the UI — are all plugins.
Key facts
| What it is | An open-source agent harness (runtime) that connects LLMs to real work: files, tools, shell commands, permissions, long-running tasks. |
|---|---|
| Core idea | Model + Harness = Agent. Everything is a plugin, powered by the Cordis plugin meta-framework. |
| Status | v0.1 developer preview released August 13, 2026, under the MIT license. Rapid iteration; breaking changes expected. |
| Four modes | Standard (everyday tasks), Minimal (benchmarking, shell + file editing only), PTC (programmatic tool calling), Creative (experiment with plugins in-memory). |
| Official links | deepseek.com/harness · github.com/deepseek-ai/deepseek-harness |
Quickstart install
Requires Node.js (22.19+ on 22.x, or 24+; check the official README). Then run:
npx @deepseek-ai/dsh webOpen http://127.0.0.1:3080 in your browser. To build from source:
git clone https://github.com/deepseek-ai/deepseek-harness.git
cd deepseek-harness
pnpm install
pnpm run build
pnpm dsh webNote: the harness itself is free (MIT), but you will still need to configure a model service for agents to execute tasks. See the Chinese install guide for troubleshooting tips.
More from this site (Chinese)
What is DeepSeek Harness?
Architecture, Model + Harness = Agent, the four modes, and core capabilities — in Chinese.
Plugin ecosystem
How the plugin system works, where to find community plugins, and how to publish your own.
vs Claude Code / Codex
Open-source plugin architecture versus polished closed-source products: an objective comparison.
FAQ
Free? MCP relationship? Model support? Production readiness? Quick answers.