Plugin-first runtime
Models, tools, skills, storage, sessions, scheduling and interfaces can be composed through plugins.
DeepSeek's official open-source agent runtime
Build and extend agents with DeepSeek Harness
Understand DeepSeek Harness (DSH), start the official developer preview, and discover selected plugins, tools and infrastructure for practical agent workflows.
Developer Preview MIT open source Everything is a Plugin
Model + Harness = Agent
What is DeepSeek Harness?
DeepSeek Harness, abbreviated DSH, is an open-source agent harness developed by DeepSeek AI. It is the runtime layer around a model: it connects model providers, tools, skills, sessions, workspaces, storage, permissions and user interfaces so an agent can complete multi-step work instead of returning an isolated answer.
The framework is built around the idea that everything is a plugin. Its Cordis-based architecture exposes major capabilities as replaceable services, which makes DSH useful for developers who need to inspect, extend or self-host an agent workflow. The current release is a developer preview, so APIs and plugin contracts can still change.
Traceable sessions
Agent work is kept as structured session state that can be inspected, continued and debugged.
Multiple workflows
Use the Web UI, CLI and SDK-oriented paths for interactive work, automation and custom products.
Local control
Run the framework locally, choose a workspace and define model, tool and permission boundaries.
Selected DSH resources
Start with first-party materials, then explore community discovery sources. Third-party entries are clearly marked and should be reviewed before installation.
DSH Official
DeepSeek Harness
Official product overview and starting point for the DeepSeek Harness developer preview.
Open resource GH Official
deepseek-ai/deepseek-harness
Official MIT-licensed source, examples, contribution history and release files.
Open resource npm Official
@deepseek-ai/dsh
Published @deepseek-ai/dsh package used to start the local Web UI.
Open resource UI Official
DSH Web UI Guide
Official instructions for configuring a model, selecting a workspace and running the first task.
Open resource DOC Official
DSH Architecture
Technical explanation of Cordis, plugins, services, session state and the agent loop.
Open resource CO Independent
Cordis
The composability framework underlying DSH and its plugin service model.
Open resource PLG Independent
dsh-plugin on GitHub
GitHub discovery feed for repositories tagged dsh-plugin; inspect ownership and maintenance before use.
Open resource DIR Independent
DeepSeek Code Directory
Independent, category-based directory for discovering DSH plugins, tools and infrastructure.
Open resourceInfrastructure around the agent
A capable model is only one part of an agent. A dependable DSH setup also needs clear runtime, data and safety boundaries.
LAYER 1
Models and providers
Configure DeepSeek or compatible providers, credentials, model routes and context policies.
LAYER 2
Tools and integrations
Connect file operations, shell commands, browsers, APIs and MCP-style services with explicit permissions.
LAYER 3
Sessions and storage
Persist context, events and workspace state so work can be resumed, audited and debugged.
LAYER 4
Sandbox and governance
Limit filesystem and command access, require approval for risky actions and keep secrets outside prompts.
Start DSH locally
Install Node.js, run the official package and open the URL printed by the command. The Web UI is served on 127.0.0.1:3080 by default.
npx @deepseek-ai/dsh web Developer preview: expect compatibility-breaking changes. Pin versions and test plugins in an isolated workspace before production use.
How we select and evaluate resources
A plugin directory is useful only when entries can be checked. Before connecting any extension to an agent workspace, review the following signals.
- Clear source repository, license and maintainer identity
- Recent maintenance and an understandable release history
- Documented permissions, network access and data handling
- Compatibility with the current DSH developer-preview version
- Reproducible setup instructions and a safe way to uninstall or disable it
DeepSeek Harness FAQ
Straight answers to common questions about DSH installation, plugins and use cases.
What is DeepSeek Harness?
DeepSeek Harness is DeepSeek AI's open-source agent runtime. It surrounds a model with tools, sessions, workspaces, storage, permissions and interfaces needed to run multi-step agent tasks.
Is DeepSeek Harness officially released?
Yes. DeepSeek publishes the official product page, repository and npm package, but the project is currently labeled Developer Preview rather than a stable production release.
How do I install DeepSeek Harness?
With Node.js installed, run npx @deepseek-ai/dsh web. The command starts a local Web UI and prints its address. You can also clone the official repository to develop from source.
What are DSH plugins?
Plugins provide focused capabilities such as model access, tools, skills, storage, scheduling, interfaces and integrations. Because DSH is plugin-first, these parts can be composed or replaced.
Can I use community plugins safely?
Treat every third-party plugin as executable code. Review its source, license, permissions, network behavior, maintenance and version compatibility, then test it in an isolated workspace.
Explore the DeepSeek Harness ecosystem
Begin with the official runtime and documentation, then add only the plugins and infrastructure your workflow actually needs.