DeepSeek Harness (dsh) is an open-source agent harness developed by DeepSeek AI that uses a plugin-first architecture. It's built on top of the Cordis framework and provides a flexible foundation for building AI agents with composability and extensibility at its core.

The "Everything is a Plugin" architecture means that every component in DeepSeek Harness is designed as a plugin. This includes core functionality, extensions, and even the runtime environment. This approach enables unprecedented flexibility, allowing developers to compose, replace, or extend any part of the system without modifying the core codebase.

Getting started is simple. You can either run it directly from npm or clone from source:

From npm:

From source:

git clone https://github.com/deepseek-ai/deepseek-harness.git
cd deepseek-harness
pnpm install
pnpm run build
pnpm dsh web

The command starts the Web UI at http://127.0.0.1:3080 by default.

DeepSeek Harness is currently in developer preview and is iterating rapidly. This means the API and functionality may change frequently, and there might be compatibility-breaking changes. We recommend checking the documentation and GitHub discussions for the latest updates.

DeepSeek Harness is licensed under the MIT License. This is a permissive open-source license that allows you to use, modify, and distribute the software freely. You can find the full license text in the LICENSE file in the repository.

We welcome contributions! You can contribute by:

• Submitting bug reports and feature requests via GitHub Discussions
• Contributing code to the repository - see the CONTRIBUTING.md guide
• Creating plugins and sharing them with the dsh-plugin topic
• Helping with documentation and examples

You can join our community in several ways:

• GitHub Discussions: For feedback, questions, and ideas
• Discord Community: Real-time discussions with other developers
• Plugin Discovery: Add the dsh-plugin topic to your plugin repositories

Links to all community resources are available on the main page.

Cordis is the framework that powers DeepSeek Harness. It's designed for spatiotemporal composability, as described in the research paper _A Programming Paradigm for Spatiotemporal Composability_. Cordis provides the foundational architecture that enables DeepSeek Harness's plugin-first approach.

DeepSeek Harness requires:

• Node.js: Version 18.x or higher
• Memory: At least 4GB RAM recommended
• Storage: Approximately 500MB for installation
• Operating System: Windows, macOS, or Linux

The Web UI runs in any modern browser with JavaScript enabled.

Creating a plugin involves:

• Understanding the plugin architecture and interfaces
• Implementing the plugin lifecycle methods
• Defining plugin metadata and dependencies
• Testing your plugin with the harness
• Sharing your plugin with the dsh-plugin topic

Detailed documentation for plugin development is available in the repository.

Bugs and feature requests should be submitted through:

• GitHub Issues: For bugs and technical issues
• GitHub Discussions: For general questions and feature suggestions

When reporting bugs, please include:

• Steps to reproduce the issue
• Expected behavior
• Actual behavior
• Environment details (OS, Node.js version, etc.)

Our roadmap includes:

• Enhanced plugin ecosystem with marketplace
• Improved developer tools and debugging capabilities
• Expanded documentation and examples
• Performance optimizations
• Additional language support beyond TypeScript
• Enterprise features and support options

Follow our GitHub repository and announcements for the latest updates.