DeepSeek Harness Quickstart: Boot the Web UI, Then Close One Repo Loop

What this tutorial is for

DeepSeek Harness (CLI name dsh) is DeepSeek AI's open-source agent runtime. It is not chat.deepseek.com and it is not a new model. It is the layer around a model: tools, files, sessions, a loop, a UI. The official README is explicit: this is a developer preview and there will be compatibility-breaking changes.

The useful learning order is not to finish the Cordis paper first. Close one small loop.

Step 1: Boot the Web UI from npm

Install Node.js, then:

sh npx @deepseek-ai/dsh web

That starts the Web UI at http://127.0.0.1:3080 by default. The invoking directory is the default filesystem location, but a fresh Web UI has no selected workspace until you add one.

From source:

sh git clone https://github.com/deepseek-ai/deepseek-harness.git cd deepseek-harness pnpm install pnpm run build pnpm dsh web

Commands and the default port come from the official README and Web UI guide. Versions change; follow the current repo docs before you deploy.

Step 2: Add a model key, then a workspace

Open Settings → Models, enter a DeepSeek API key and save. Official note: keys are write-only and live in $DSH_HOME/.credentials.yaml. Model changes apply on the next request without a restart.

Then click Choose workspace, add the project directory where you started dsh, and select it. The session composer stays unavailable until a workspace is selected.

Step 3: Send one short task

The first example in the official Web UI guide is enough:

Summarize this repository and identify its main packages.

Do not ask it to edit code, push a repo or install dependencies yet. Watch whether it can read files and run commands under the active permission policy. The Web UI asks before operations that need approval.

Two paths to skip on day one

• headless: dsh --profile headless "job" runs one persisted session, prints the answer and exits. Good for scripts, bad for learning the UI.
• Custom providers: besides a DeepSeek key, the official guide covers Anthropic / OpenAI catalog providers and custom OpenAI-compatible endpoints. Get DeepSeek working first, then swap endpoints.

Boundaries

Do not treat a developer preview as a stable production base. Use it to learn the architecture, write plugins and experiment. Do not pin core business flows to the current command and config. The license is MIT.

Hub: /en/deepseek-harness.

FAQ

How do I install DeepSeek Harness?

Install Node.js and run npx @deepseek-ai/dsh web. From source: clone, then pnpm install && pnpm run build && pnpm dsh web.

Why is the composer disabled?

The official Web UI guide says the session composer stays unavailable until a workspace is selected. Choose a workspace first.

Does it only work with DeepSeek models?

No. The official providers guide also covers catalog providers and custom OpenAI-compatible endpoints.

Official source: DeepSeek Harness README and Web UI guide. This is a learning path, not a replacement for the official docs.