DEV Community

Antonne Dillard
Antonne Dillard

Posted on • Originally published at teamorouter.com

The Ultimate Guide to DeepSeek Harness (DSH): What It Is and How to Install

#deepseek #ai #agents #devtools

DeepSeek just shipped an open-source agent harness — and you can run it with one command:

npx @deepseek-ai/dsh web
Enter fullscreen mode Exit fullscreen mode

That pulls the package and starts a Web UI at http://127.0.0.1:3080. No clone, no build.

What is DeepSeek Harness (dsh)?

DeepSeek Harness (dsh) is DeepSeek AI's open-source agent harness — the framework that turns a language model into a worker that can actually edit code and run commands. Built on an "everything is a plugin" architecture, powered by the Cordis runtime, and released under the MIT license. It's in developer preview, so interfaces will still shift — but for solo devs and cheap agent experiments it's fully usable today.

The industry has converged on a clean formula:

Agent = Model + Harness
Enter fullscreen mode Exit fullscreen mode

The model is the brain — the weights that predict tokens. The harness is everything else: the tools the agent can call, the filesystem and shell it can touch, how sub-agents pass context, and when execution stops. A bare model is bad at memory and tool use on its own; the harness turns a chat model into a worker.

What model does it run?

dsh is model-agnostic, but its natural fit is the DeepSeek V4 family:

  • DeepSeek V4 Pro — the flagship reasoning model, up to 1M context, multiple reasoning-effort levels.
  • DeepSeek V4 Flash — the fast, cheap sibling for high-frequency work.

To route dsh through a free tier, set two env vars:

export DEEPSEEK_API_KEY="sk-teamo-your-key"
export DEEPSEEK_BASE_URL="https://api.teamorouter.com/v1"
Enter fullscreen mode Exit fullscreen mode

Then npx @deepseek-ai/dsh web again. That's the whole setup — dsh speaks the OpenAI protocol, so any OpenAI-compatible endpoint works.

What's in the Web UI

http://127.0.0.1:3080 is the front-end of an agent runtime:

  • Sessions — create, switch, rename, run several in parallel.
  • Model selection — swap deepseek-v4-pro vs deepseek-v4-flash per session.
  • Tool panel — the tools the agent can call: bash, file read/write, sub-agents, web search.
  • Goals — break long tasks into stateful goals, pause / resume.

Don't treat it as "another chat window" — you give a goal, the agent loops over tools until done, and the UI is your observation deck.

Prove it works

Give the agent this task:

List the current directory, find the README, summarize its first
paragraph in one sentence, and write it to /tmp/summary.txt
Enter fullscreen mode Exit fullscreen mode

dsh calls bash and the file tools in sequence, then reports back. That's the whole idea — the harness turns a model into a worker.

Run your first agent loop → Get a free key on TeamoRouter

Top comments (0)