sakutto
Generative AI

How to Use DeepSeek Harness: The MIT-Licensed Open-Source Agent Foundation

DeepSeekAgentOpen source
How to Use DeepSeek Harness: The MIT-Licensed Open-Source Agent Foundation

What DeepSeek Harness is

DeepSeek Harness at a glance

Command
dsh
License
MIT
Foundation
Cordis (a plugin composition framework)
Status
Developer preview (v0.1)

DeepSeek Harness is an open-source agent foundation released by DeepSeek AI.

The everything-is-a-plugin design

Its character comes down to a single idea: everything is a plugin. The project released v0.1 as a developer preview on August 13, 2026, stating that models, tools, skills, sessions, sandboxes, filesystems, loops, orchestration, and the UI are all implemented as plugins. Replacing or extending any of them is a matter of recomposition rather than modifying the core.

View official source →
"🧩 DeepSeek Harness v0.1 is now available in Developer Preview!"/"We’re opening it up to developers building agent harnesses worldwide and open-sourcing the codebase in MIT license."/"Powered by the Cordis meta-framework, DeepSeek Harness is an agent harness built around one core idea: Everything is a plugin. Models, tools, skills, sessions, sandboxes, filesystems, loops, orchestration, and UI are ALL implemented as plugins, and can be mixed, matched, replaced, and extended." — from the announcement by DeepSeek's official account

No privileged core to patch

Looking a little deeper makes the design philosophy clearer. The official documentation says there is no privileged core to patch, and that you add functionality by mounting a plugin beside the others. The model adapter, the tool registry, the session log, and the agent loop itself are all plugins, replaceable from configuration.

View official source →
"Every part of the product is a plugin, including the model adapter, the tool registry, the session log, and the agent loop itself, so every part is replaceable from configuration."(from the Cordis section)/"There is no privileged core to patch: you extend dsh by mounting a plugin beside the others, and registrations are effects that unwind when their plugin unloads."(from the next paragraph of the same section) — from the DeepSeek Harness official documentation

How to use DeepSeek Harness

Two ways to start it

Via npm
Install Node.js and start with a single npx
From source
Clone and build with pnpm

Two launch paths are provided. If you only want to try it, the npm route is faster.

The shortest path with npx

As long as Node.js is installed, the following single line brings up the Web UI. The default address is http://127.0.0.1:3080.

npx @deepseek-ai/dsh web

The Node.js version requirement is ^22.19.0 || >=24.0.0. It will not start on an older Node, so checking that first saves you a detour.

View official source →
"Install `Node.js`, then run:"/"npx @deepseek-ai/dsh web"/"The command starts the Web UI, served at `http://127.0.0.1:3080` by default."(all from the Run from `npm` section) — from the DeepSeek Harness README
View official source →
"engines": { "node": "^22.19.0 || >=24.0.0" } — from package.json at the repository root

Running from source

If you want to work on the internals, fetch the repository. The package manager is pnpm: clone, install, build, then start.

git clone https://github.com/deepseek-ai/deepseek-harness.git
cd deepseek-harness
pnpm install
pnpm run build
pnpm dsh web
View official source →
"To run from a repository checkout:"/"git clone https://github.com/deepseek-ai/deepseek-harness.git"/"cd deepseek-harness"/"pnpm install"/"pnpm run build"/"pnpm dsh web"(all from the Run from source section) — from the DeepSeek Harness README

How it is composed, and what to watch for

Units of composition

Profile
A named composition listing the bundles it stacks. web and headless ship as standard
Bundle
A distribution unit bundling config rows and the code they mount

Two units make that replaceability work: profiles and bundles.

Profiles and bundles

What gets assembled at boot is a layered plugin tree. A profile is a named composition that lists the bundles it stacks, and web and headless ship as templates. A bundle is a distribution format for config rows and the code they mount, and it stays patchable by the layers above it. You can check what is actually running in your environment with the following command.

dsh --profile web --dump-config
View official source →
"A running `dsh` is a plugin tree composed at boot from ordered layers."(from the Profiles and bundles section)/"A profile is a named composition stored in the Harness home. It lists the bundles it stacks … `web` and `headless` ship as templates."(same section)/"A bundle is a distribution format for Cordis config rows and the code they mount, so whatever it inserts stays patchable by the layers above it."(same section)/"dsh --profile web --dump-config"(the command example in the same section) — from the DeepSeek Harness official documentation

What to keep in mind

A design that is free to swap is, flipped around, one whose behavior shifts substantially with configuration. Given that compatibility-breaking changes will land during the preview, trying it in an isolated environment is more realistic than wiring it into daily production tooling. For the model side, see the DeepSeek model articles.

When you want to read English documentation on your own machine as markdown with its headings and tables intact, the following tool can help.

Free ToolURL to Markdown ConverterConvert any public web page URL to Markdown. Preserves headings, tables, lists, and links — perfect for LLM and RAG preprocessing, research notes, and archiving web articles.Try it now →

FAQ

Q. What is DeepSeek Harness?
It is an open-source agent harness (the foundation that runs an agent) developed by DeepSeek AI. The command is dsh, it follows an everything-is-a-plugin design, and it is built on a framework called Cordis.
DeepSeek Harness Official Repository (README)
DeepSeek Harness (`dsh`) is an open-source agent harness developed by DeepSeek AI. DeepSeek Harness Official Repository (README)
Q. What license is DeepSeek Harness under?
The MIT license. Third-party dependency licenses are disclosed in THIRD_PARTY_NOTICES.md. It is among the easier licenses to work with, including for commercial use.
DeepSeek Official Account (August 13, 2026)
We’re opening it up to developers building agent harnesses worldwide and open-sourcing the codebase in MIT license. DeepSeek Official Account (August 13, 2026)
Q. Can I use DeepSeek Harness in production right now?
It is not recommended. The project states that it is in developer preview and that compatibility-breaking changes will happen. Starting with trials and evaluation is the safer path.
DeepSeek Harness Official Repository (README)
DeepSeek Harness is currently in developer preview and is iterating rapidly. THERE WILL BE COMPATIBILITY-BREAKING CHANGES. DeepSeek Harness Official Repository (README)

Related Tools

Related Tool Categories

Articles