Install DeepSeek Harness

After this tutorial you will

• Have DeepSeek Harness running on your own computer
• Have its web UI open in your browser
• Know where it stores data and how to remove it completely

The whole thing takes about ten minutes. The only prerequisite is being able to paste a command into a terminal — every step is copy-paste ready.

Step 1: Install Node.js

DeepSeek Harness is written in Node.js and needs Node.js 22 or newer (grab the latest LTS release).

First check whether you already have it. Open a terminal (Terminal on macOS, PowerShell on Windows) and type:

node -v

• If it prints v22.x.x or higher (e.g. v24.x.x) — great, skip ahead.
• If the command is not found or the version is below 22, download the LTS installer from nodejs.org, click through it, then reopen the terminal and try node -v again.

Fun fact: Node.js is like a power strip — dsh is the appliance that plugs into it. Install it once and forget about it.

Step 2: Launch with one command

Paste this into the terminal and press Enter:

npx @deepseek-ai/dsh web

What happens on the first run:

1. npx downloads DeepSeek Harness from npm automatically (a minute or two, depending on your connection);
2. It starts automatically and prints an address like:

DeepSeek Harness is running at http://127.0.0.1:3080

3. Open your browser at http://127.0.0.1:3080 — there is the UI.

Keep that terminal window open. It is the engine of DSH; closing the window shuts DSH down. To stop the service, press Ctrl + C in the terminal.

npx is Node.js's "run-on-demand" tool: it downloads and runs a package for you without a permanent install. Perfect for trying things out.

Step 3: Pick a workspace

On first launch the chat input is disabled — DSH doesn't yet know which folder it may work on.

Click Choose workspace, add the folder you want it to work in (your code project, for example), then select it. The input unlocks once a workspace is selected.

The workspace is DSH's sandbox: every file it reads or writes stays inside that folder. To switch projects, click Choose workspace again at any time.

Step 4 (optional): Add an API key so it can talk

At this point DSH still can't answer anything, because no model is attached yet. This takes two minutes:

1. Sign up at platform.deepseek.com and create an API key;
2. In DSH, open Settings → Models, paste the key, and save.

For the full walkthrough (top-up, common errors), see the model configuration tutorial. After this step DSH is fully usable.

Optional: install dsh globally

With npx you type a long command every time. For daily use, a global install is nicer:

npm install -g @deepseek-ai/dsh

From then on, starting is just:

dsh web

Both ways are functionally identical — one just has a shorter command. Starting with npx is perfectly fine; you can switch anytime.

Where the data lives

Everything DeepSeek Harness stores sits in one directory:

It mainly contains:

• profiles/ — the configuration and installed plugins of each profile;
• .credentials.yaml — the API keys you saved;
• session logs — the full transcripts of your conversations.

To back up: copy the whole directory. To factory reset: quit dsh, delete the directory, and the next launch starts fresh.

Quick troubleshooting

• command not found: npx: Node.js isn't installed properly — back to step 1, and reopen the terminal after installing.
• npx download is very slow: it's the network. You can switch to a faster registry: npm config set registry https://registry.npmmirror.com, then retry.
• Browser can't reach port 3080: make sure the terminal window is still open and error-free. For port conflicts and more, see the FAQ.
• Antivirus/firewall prompt: dsh opens a local port to serve the web UI; choose "allow". It does not expose anything to the internet.

Next up

Now that it runs, wire up a model so DSH gets its brain:

Configure a model (API Key) →