Developer preview · MIT

DeepSeek Harness
Everything is a plugin · Cordis kernel

Agent = Model + Harness. This is DeepSeek’s own shell, not a random pip adapter. Standard / Code / Minimal / Creator modes. Append-only trajectory logs.

#npx dsh web#Cordis#4 modes#MIT preview

Highlights

npx

@deepseek-ai/dsh web

Serves Web UI at 127.0.0.1:3080. Or clone + pnpm.

Cordis

Plugin kernel

github.com/cordiverse/cordis — mount, deps, events.

Minimal

Official bench mode

bash + str_replace_editor only. That’s the 0731/0813 footnote.

break

Preview warning

README: THERE WILL BE COMPATIBILITY-BREAKING CHANGES.

What dsh is

May hiring said Model + Harness = Agent and named Claude Code / Cursor / Codex as the layer to fill. August’s developer preview is that line shipping.

Why it is trending

Do note the distinction from the third-party "deepseek-harness" protocol wrappers on PyPI: the official project lives under the deepseek-ai GitHub organisation in a repository named deepseek-harness. Community plugins are usually tagged with the dsh-plugin topic.

Timeline

2026-05

Public Code Harness hiring.

2026-07-31

Flash 0731 changelog names upcoming Harness Minimal.

2026-08

dsh developer preview + Cordis site.

Confirmed vs caution

Confirmed

Official site + GitHub README + four modes + MIT + bench footnote.

Caution

Three expectations to set: dsh is a developer preview and its APIs will change in breaking ways; its feature scope is not equivalent to Claude Code; and while the software is free under MIT, model tokens still cost money.

How to choose

Need official DeepSeek scores

Use Minimal + max + t=1.0 p=0.95 as they documented.

Need a daily coding agent

Standard/Code modes, knowing it’s preview. Or Claude Code / OpenCode.

How to use it

npx @deepseek-ai/dsh web · git clone github.com/deepseek-ai/deepseek-harness

On QCode

dsh runs on your own key. QCode currently lists deepseek-v4-flash and deepseek-v4-pro on /models, and those can be pointed at dsh like any other OpenAI- or Anthropic-compatible endpoint.

FAQ

Official?

Yes, deepseek-ai.

What is Cordis?

The plugin kernel under dsh.

Why Minimal?

To score the model, not the kitchen sink.

Paid?

MIT software. You pay model APIs.

Replace Claude Code?

Same category, preview maturity.

Site harness-engineering-guide?

The on-site /harness-engineering-guide covers the generic concept of a harness — the scheduling, tooling and context layer that sits around a model. This page covers dsh, DeepSeek's own concrete implementation of that idea.

Sources

deepseek.com/harness/en/ ; github.com/deepseek-ai/deepseek-harness ; 0731/0813 changelogs.

Concept page already exists — this is the product

Link /harness-engineering-guide. Put Cordis in an H2 for that exact query.

Related

Not affiliated beyond citing official repos. Preview software.