DeepSeek Harness: what it is, what it costs and how to install it

DeepSeek Harness, also called dsh, is an open-source agent harness from DeepSeek. A harness is the program around a model that lets it work on a project: it reads and edits files, runs commands and keeps a plan. DeepSeek Harness is free (MIT license). You pay only for the DeepSeek API calls it makes. The GitHub repository calls it a developer preview.

What DeepSeek Harness is

According to the repository README, DeepSeek Harness is built on an "everything is a plugin" architecture and is powered by Cordis, a TypeScript plugin framework. In other words, the parts of the agent are plugins that can be swapped. The README lists the topics ai-agents, cordis, dsh and dsh-plugin.

The Web UI guide says the agent can read and edit files in your workspace, run commands, delegate work and maintain a plan. It asks for approval before operations that need it under the active permission policy.

DeepSeek's API docs list Harness under "Agent Integrations," next to tools such as Claude Code and Codex. They describe it as "now in developer preview for agent harness developers worldwide."

What it costs

The software costs nothing. The repository is MIT licensed. What you spend is API usage, billed by the model you pick. DeepSeek's pricing page lists two models, both with a 1M-token context window. Prices are per million tokens in US dollars, as of November 6, 2026.

Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays. Off-peak prices are half the peak prices. The page also says Search and Fetch are billed at $0.75 per 1,000 successful requests each. The docs do not say whether Harness uses those tools.

Agents send long prompts again and again, so the cache-hit price matters most. A real session's cost depends on how much your project and conversation repeat between calls. The docs give no per-session estimate.

How to get it

You need Node.js and a DeepSeek API key from platform.deepseek.com. Then run:

npx @deepseek-ai/dsh web

This starts the Web UI at http://127.0.0.1:3080 and opens it in your browser. Add --no-open to skip the browser. To build from source you need pnpm:

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

These commands come from the README.

Once it is running:

1. Open Settings, then Models, enter your API key and save. The guide says the model route works immediately, without a restart.
2. Click Choose workspace, add your project directory and select it. The session composer stays unavailable until a workspace is selected.

The dsh process uses the directory it was started in as its default filesystem location. Other providers and custom OpenAI-compatible endpoints are covered in the project's model configuration guide.

Risks and what is not settled

The README warns: "THERE WILL BE COMPATIBILITY-BREAKING CHANGES." It also tells users to read SAFETY.md before running the project. An agent that edits files and runs commands should be pointed at a project under version control.

The repository shows no releases at the time of writing. Its page showed about 274,000 stars and a last update on November 4, 2026. Third-party sites quote very different star counts and user numbers,. Pandaily reported a figure of 1.54 million daily users and a coming plugin marketplace. DeepSeek's pages do not say this, though the docs link a plugin market guide.

The Web UI guide covers one mode. It links to a CLI README for other modes.