DeepSeek Harness · Get Started

Launch a Local Web UI in Minutes Or Install Fully from Source

Official recommendation: after installing Node.js, run npx @deepseek-ai/dsh web. The source path is clone the repo, pnpm install / build, then pnpm dsh web. If you would rather confirm prompts first, start with DeepSeek-V4 on the DeepSeek web version.

npm

npx Launch

@deepseek-ai/dsh web

git

Clone Source

deepseek-ai/deepseek-harness

Preview

Breaking Changes

Do not blindly pin production versions

Web

DeepSeek Web Version

Validate the task with zero install

Recommended Path

Validate on the DeepSeek Web Version + Execute with Local Harness

Every Use button on this site still jumps to DeepSeek web version chat, so you are not stuck in a preview local environment with nowhere to ask the model.

1. Spell Out the Task on the Web Version

Open the DeepSeek web version and use DeepSeek-V4 to write the goal, constraints, and success criteria. This step needs zero install and fits anyone searching for the DeepSeek web version.

DeepSeek web version

2. Install Node.js

Running dsh locally requires Node.js. Prepare it, then run the official npx command—avoid cloning on a machine that does not have the runtime yet.

DeepSeek Harness Node.js

3. Start the Web UI with npx

Run npx @deepseek-ai/dsh web and, per official docs, visit 127.0.0.1:3080 by default. This is the fastest way to see the Harness interface.

npx dsh web

4. Go to Source When You Need to Change the Framework

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

DeepSeek Harness source

5. Send Feedback via Discussions

During preview, submit feedback or bugs through GitHub Discussions rather than assuming Issue templates are already stable.

DeepSeek Harness feedback

6. Tag Plugins with dsh-plugin

When publishing a community plugin, add the GitHub topic “dsh-plugin” so other DeepSeek Harness developers can find it.

dsh-plugin
Official Commands

Two Launch Paths Matching the README

Treat the current GitHub repository README as source of truth; the commands below are for this intro page so they are not mixed up with outdated blogs.

Run via npm / npx

Run after installing Node.js to start the Web UI.

npx @deepseek-ai/dsh web

Run from Source

After clone, install and build following the repository instructions. See the official README for the full four steps.

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

Explore the Ceiling of Intelligence with Harness Developers Worldwide

The DeepSeek Harness developer preview is still in testing; core plugins and the base API will keep iterating. We gather official entry points so you can jump upstream from this intro, while this site’s use path still returns to the DeepSeek web version.

GitHub Repository

deepseek-ai/deepseek-harness: source, architecture docs, developer guide, and license.

Discussions

Official recommendation is to submit feedback and bugs via Discussions—a better fit for preview-stage communication.

dsh-plugin topic

Add that topic to your plugin repository to enter the discoverable community plugin list.

DeepSeek Web Version

If you do not want to set up a local environment first, keep designing the Agent task clearly with the DeepSeek-V4 web version.

Chat with DeepSeek-V4 Once before Installing

Write down the goal and constraints, then decide whether to launch DeepSeek Harness locally.

Getting-Started FAQ

Common Sticking Points When Installing DeepSeek Harness

Ports, preview risk, and why this site still jumps to the DeepSeek web version first.

01 Browser will not open after npx?

Confirm Node.js is installed, the command is npx @deepseek-ai/dsh web, and you visit the official default address http://127.0.0.1:3080. If the port conflicts, check the current repository Web UI guide.

02 What are the source-install dependencies?

The official README specifies pnpm install, pnpm run build, then pnpm dsh web. Exact Node / pnpm versions follow the repository.

03 Why do buttons go to the DeepSeek web version instead of GitHub?

This site exists to help people use DeepSeek / the DeepSeek web version / DeepSeek-V4. The Harness intro is for understanding the Agent runtime; actually getting started still begins with web-version chat, then local install as needed.

04 Can I connect non-DeepSeek models?

Architecturally the model is a plugin, and community notes mention connecting other models via a custom provider. Environment variables and configuration follow official docs and the current preview.

05 Should I put it in production now?

Do not pin the preview as a stable runtime. Official guidance warns about breaking changes in capital letters. For production conversation and plan validation, prefer the DeepSeek web version and the official API.

06 How does the Chinese-speaking community communicate?

The Chinese README mentions WeCom groups, official accounts, and similar entry points—follow the current official repository notes. The international community also has channels such as Discord.

Local Environment Ready? Still Start from the Web Version

Use entry points on this site all jump to the DeepSeek web version, so you always have a working DeepSeek conversation.