DeepSeek Agent Starts with Harness

Download and launch DeepSeek Harness (DSH), configure a model and workspace, extend it with plugins, and run your first Agent.

DeepSeek Harness Setup GuideBrowse all Harness guides

Prefer a bundled runtime? Download DSH Desktop →

Independent DeepSeek AI Agent guide — not affiliated with or endorsed by DeepSeek.

What is DeepSeek Harness?

DeepSeek Harness (DSH) is DeepSeek's official open-source agent harness. It is not a new model; it is the runtime that helps models understand environments, use tools, and keep working on tasks.

Models, tools, skills, sessions, sandboxes, storage, agent loops, and the UI are all plugins that can be swapped and recomposed.

View setup guideBrowse all Harness guides
npm latestDeepSeek Harness v0.1.5-rc.2 →npm alphaDeepSeek Harness v0.1.7-alpha.2 →
Terminal — dsh

$ npx @deepseek-ai/dsh web

How to Use DeepSeek Harness

Launch the DeepSeek Harness Web UI, configure a model and API key, then choose a workspace and runtime mode for your first task.

  1. Step 1

    Launch DeepSeek Harness

    With Node.js installed, run npx @deepseek-ai/dsh web and the Web UI opens in your browser. Use the community DSH Desktop installer if you prefer a bundled runtime.

  2. Step 2

    Configure a model and API key

    Open Models, add a provider, enter its API key, and choose a model. Use a built-in provider or add a custom service with its base URL, protocol, and model ID.

  3. Step 3

    Choose a workspace and run a task

    Create a session, choose a workspace, model, and Standard, Code, Minimal, or Creator mode, then enter your task. Install plugins and compose a Profile when you need more capabilities.

DeepSeek Harness Plugins and Themes

Once Harness is running, extend its capabilities with plugins and personalize its interface with installable themes.

Plugin Market

Plugin Market

Search more than 3,300 community plugins and copy their installation commands. →

Theme Gallery

Theme Gallery

Preview installable themes and choose the look that fits your workspace. →

See what people are building with DeepSeek Harness

Explore reviewed Agents and presets built on DeepSeek Harness—from research and writing to engineering and specialist workflows.

Browse all native Agents

Keep learning after your first run

Follow the current candidate, V4.1 Flash 1017, and DeepSeek Search/Fetch.

Browse all Harness guides

Already using another Agent?

Using Codex, Claude Code, OpenCode, or another Agent? Find the API endpoint, model name, and setup steps for your tool.

See Agents that work with the DeepSeek API →

DeepSeek Harness questions

The essentials before installing, extending, or building with DSH.

What is DeepSeek Harness?

+
DeepSeek Harness is DeepSeek's official open-source Agent runtime. It provides sessions, tools, workspaces, sandboxing, plugins, Agent loops, and Web and terminal interfaces around a model.

Is DeepSeek Harness a model?

+
No. Harness is the runtime that connects a model to tools and an execution environment. You configure the model provider and API key separately.

How do I install it?

+
With Node.js installed, run npx @deepseek-ai/dsh web. The community DSH Desktop bundles the runtime for users who prefer not to configure Node.js locally.

What can plugins and presets change?

+
Plugins provide models, tools, storage, interfaces, sandboxes, and other capabilities. Presets compose those parts into a ready-to-use Agent for a particular workflow.

Where can I find Agents built for DSH?

+
The native Agent directory lists reviewed projects with concrete DeepSeek Harness presets, bundles, or Agent structures, grouped by research, creative, engineering, and specialist use cases.