DeepSeekPlugins
GitHub
Independent community directory · Updated August 2026

DeepSeek Harness plugins, without the guesswork.

One place to find useful community extensions for dsh. We link to the source, show the install command, and tell you what each plugin actually changes.

See the directory Run dsh
Start the official harnessNode.js required
$ npx @deepseek-ai/dsh web

Then choose a plugin, check its repository, and add it to the profile you actually use.

Why this directory exists

The useful part is knowing what you are installing.

DeepSeek Harness is still in developer preview, and its plugin ecosystem is moving faster than a polished catalog can keep up. Interesting extensions appear on GitHub before they have documentation, release notes, or a familiar package name. That makes discovery easy and evaluation hard. DeepSeekPlugins is a working index for developers who want to try the ecosystem without opening dozens of tabs first. Start with the complete plugin directory, narrow the list by capability, and open a project page for the source, license, maintainer, and copyable install command.

This is an independent community resource, not an official DeepSeek product and not a security seal. “Verified” here means that we checked the public repository and found a documented connection to dsh; it does not mean we audited every line of code. Read the maintainer's README, review permissions, and test unfamiliar extensions in a disposable profile. If you are new to the harness, follow the installation guide. If you want to extend it yourself, the first-plugin tutorial explains the moving parts without pretending the preview APIs are settled.

Before you install

Four checks that save time later

The ecosystem is young. A repository name and a star count do not tell you whether an extension belongs in an important workspace. These four checks do.

Search the full directory →
01

Start with the capability gap

Choose one missing capability rather than installing a bundle of unrelated extensions. A vision tool helps a text-only model inspect screenshots; a session plugin improves navigation or recovery; a provider plugin changes how the harness reaches a model. That narrow goal makes compatibility checks and rollback much easier.

02

Read the source and manifest

Open the linked GitHub repository before installation. Confirm that the package declares a dsh bundle or documents an equivalent source-based setup. Review permissions, filesystem access, network calls, build scripts, configuration, and the license. A GitHub topic is a discovery signal, not proof that code is safe.

03

Use the maintainer’s command

DeepSeek Harness plugins can come from npm, GitHub, a tarball, or a local checkout. Copy the command shown on the plugin page, but compare it with the repository README first. Install into the intended profile, restart when the bundle list changes, and pin a release or commit when reproducibility matters.

04

Test in a disposable profile

Developer-preview APIs can change quickly. Try unfamiliar plugins in a separate profile and an unimportant workspace. Confirm that tools register correctly, UI elements load, cleanup works, and removal leaves the profile healthy. Only move an extension into important sessions after you understand its behavior and failure modes.

Pick by capability

Start with what you need

A model adapter solves a different problem from a session tool. Use these categories to find one missing piece instead of installing a bundle you cannot explain later.

View all plugins →
01M
ModelsModel providers, routing, and inference adapters.
102T
ToolsNew actions and structured tools for agents.
1203S
SkillsReusable workflows and specialist instructions.
504↺
SessionsConversation lifecycle, rewind, and orchestration.
605□
SandboxesIsolated environments for safe agent execution.
306D
StorageMemory, artifacts, and persistent state.
207U
UIWeb, terminal, and in-conversation interfaces.
12
The architecture
01/07

Core concept

What is DeepSeek Harness?

DeepSeek Harness, or dsh, is an open-source agent harness from DeepSeek AI built around a deliberately simple idea: every capability is a plugin.

Instead of locking models, tools, skills, session handling, storage, sandboxes, and interfaces into one application, dsh composes them as replaceable layers. That all-plugins architecture lets developers build a lean benchmark runner, a full coding agent, or a specialized workflow from the same foundation—and swap one part without rewriting the rest. The separation also makes experiments easier to compare because a developer can change one capability while keeping the surrounding profile stable.

The project is powered by the Cordis meta-framework and is currently in developer preview, which means the architecture is moving quickly and compatibility-breaking changes are expected. The official command-line package is @deepseek-ai/dsh. DeepSeekPlugins tracks the emerging ecosystem so developers can find community projects by capability, inspect their source, and follow the install path supplied by each maintainer.

Models+Tools+Skills+Runtime=Your harness
Start in under a minute

Run DeepSeek Harness

Install Node.js, then launch the official web experience. npm downloads the current package and dsh serves the UI locally.

Terminal$ npx @deepseek-ai/dsh web
Learn the ecosystem

Latest tutorials

Practical, source-linked guides for running, extending, and choosing dsh plugins.

Browse all tutorials →
Getting started01

How to Install DeepSeek Harness on Windows, macOS & Linux

Launch the official web UI, understand profiles, and verify your first local dsh session.

6 min readRead guide →
Development02

How to Build Your First dsh Plugin

A practical tour of bundle layers, plugin manifests, services, and local profile installation.

12 min readRead guide →
Roundup03

Top 10 DeepSeek Harness Plugins in 2026

The community extensions worth watching across vision, tools, sessions, and interfaces.

8 min readRead guide →
Plain answers

DeepSeek Harness FAQ

The short version of the questions that come up before a first dsh profile.

01What is DeepSeek Harness?

DeepSeek Harness is an open-source agent harness developed by DeepSeek AI. It is powered by the Cordis meta-framework and uses an architecture where every capability—from models and tools to storage and interfaces—is a composable plugin.

02Is DeepSeek Harness free?

Yes. DeepSeek Harness is open source under the MIT license. Individual community plugins may use different licenses, so check each plugin repository before using it.

03How do I install DeepSeek Harness plugins?

Use the dsh plugin command for the profile you want to extend. The exact package or Git source varies by plugin; every directory card links to its source and shows a copyable command.

04What is dsh?

dsh is the command-line interface for DeepSeek Harness. You can start the official web experience with npx @deepseek-ai/dsh web and manage profile plugins with the dsh plugin subcommand.