GUIDES · BEGINNER
Install DeepSeek Harness and run your first task
A verified quickstart for launching the DeepSeek Harness web UI, connecting a model, choosing a workspace, and completing a first coding task safely.
Guides articles
- Install DeepSeek Harness and run your first task
- Configure a model provider in DeepSeek Harness
- DeepSeek Harness glossary: scope, goals, turns, and seams
- DeepSeek Harness architecture: plugins, profiles, and events
- DeepSeek Harness Agent lifecycle explained
- DeepSeek Harness tool catalog: shell, files, agents, web
- DeepSeek Harness API Gateway and remote method calls
- DeepSeek Harness development setup and contributor workflow
DeepSeek Harness (dsh) is an open-source coding-agent harness from DeepSeek. The fastest way to try it is the browser-based interface included in the package. This guide keeps the first run deliberately small: launch the UI, connect a model, select a disposable workspace, and ask the agent to inspect before it edits.
DeepSeek Harness is currently described upstream as a developer preview. Use a test repository first, review tool calls, and keep secrets outside the workspace.
Prerequisites
You need a recent Node.js release and a model provider API key. Before continuing, confirm that Node and npm are available:
node --version
npm --version
This guide was checked with Node.js 24. If the launcher fails on an older release, upgrade to a current Node.js LTS release and retry.
Start the web interface
Run the official package without installing it globally:
npx @deepseek-ai/dsh web
The terminal prints the local address used by the web interface. In the current preview, it normally opens on http://127.0.0.1:3080. Keep that terminal running while you use the app.
Using npx is useful for a first evaluation because it avoids a permanent global installation. It can download the package on the first run, so review the package name before approving the prompt.
Connect a model
Open the settings area in the web interface, add a model provider, and enter the API key supplied by that provider. Select the model you want the harness to use.
DeepSeek Harness is the orchestration layer; model access and billing remain with the provider you configure. Do not paste an API key into chat messages, source files, or a repository-level environment file that may be committed.
Choose a safe workspace
Create a small test directory or use a clean Git repository. Select that directory as the workspace. The agent can read and modify files inside the scope you give it, so do not begin with a home directory or a folder containing unrelated credentials.
A useful first prompt is:
Inspect this project, explain its structure, and propose one small improvement.
Do not edit any files until I approve the plan.
This confirms that the agent can inspect the workspace while keeping the first interaction reversible.
Approve one small change
After reviewing the proposed plan, authorize a limited edit. Check the diff before accepting the result, then run the project’s existing checks. Git is your recovery layer: commit or stash your own work before granting an agent write access.
Verify the setup
Your installation is working when all of the following are true:
- the local web interface loads;
- the selected model responds;
- the harness can read the intended workspace;
- tool calls appear for review; and
- a small approved edit produces an understandable diff.
If the UI does not start, use the web UI troubleshooting checklist.
Stop or remove it
Press Ctrl+C in the terminal that started the web server. Because this quickstart uses npx, there is no required global uninstall step. Provider credentials and local application data should still be removed separately if you no longer want to keep them.
Sources and verification
The launch command and product description are based on the official DeepSeek Harness repository. Interface labels can change during the developer preview. This page records the environment and update date above so you can judge whether the instructions match your setup.
npx @deepseek-ai/dsh webDeepSeek Harness is changing quickly. Verify commands against the official repository before using them in production.
github.com/deepseek-ai/deepseek-harness