DeepSeek Harness Ships a Plugin-First Agent Runtime Where Every Capability Is Swappable

Today, the developer preview version (v0.1) of DeepSeek Harness is open for global Harness developers to test, and the source code is simultaneously open-sourced under the MIT license.

As an early preview version, many details still need improvement and polishing. The core plugins and basic interfaces will also undergo rapid iteration and evolution in the future. We look forward to hearing feedback and suggestions from Harness developers.

DeepSeek Harness adopts a design philosophy of "Everything is a Plugin." We use a plugin-based open architecture to build the Agent Harness: all Agent capabilities—including models, tools, skills, sessions, sandboxes, storage, loops, scheduling, and UI—are composed of plugins, which can be freely replaced and flexibly recombined.

Design Philosophy

• Everything is a Plugin

DeepSeek Harness is built on the Cordis plugin system, which features spatiotemporal composability. The Cordis meta-framework is only responsible for loading, unloading, and managing dependencies of plugins; all specific components of the Agent Harness are different Cordis plugins. Plugins collaborate with each other through Cordis services and events and can be freely combined at the configuration layer.

Developers can independently select, replace, or extend any capability in the form of a plugin without modifying the source code of DeepSeek Harness itself. This is the most important design principle of DeepSeek Harness: Everything is a Plugin.

• Multiple Operating Modes

For different usage scenarios, DeepSeek Harness provides four modes, each loading a different set of plugins by default:

• Standard Mode: Provides a complete set of tools;

• PTC Mode: Programmatic Tool Calling, where a piece of code generated by the model combines multiple rounds of tool calls;

• Minimal Mode: Retains only one shell tool and one file editing tool, used for model benchmarking in minimal environments;

• Creation Mode: Can inspect the current runtime, experiment with Cordis plugins in memory, and combine and create new modes based on this.

  

• Every Run is Traceable:

Everything the model sees is written to an append-only session log, including system prompts, chain of thought, tool calls and results, sub-Agent scheduling, and every context injection. In the Trajectory view, you can view this information by source. Recovery, forking, retrieval, and replay all share the same event stream.

Getting Started

• Quick Experience

On a system with the Node.js development toolchain installed, you can use the npx command to quickly start the DeepSeek Harness Web UI:

npx @deepseek-ai/dsh web

• Source Code Installation

Get the complete project source code and follow the repository instructions to complete the installation:

git clone https://github.com/deepseek-ai/deepseek-harness

Final Words

The current v0.1 version of DeepSeek Harness is just a starting point. We look forward to exploring the upper limits of intelligence together with global developers on top of an open-source, open, reusable, and composable infrastructure. We sincerely invite global Harness developers to co-build the DSH plugin ecosystem.

Some plugins created by DeepSeek Harness internal test users