DeepSeek Harness

What is DeepSeek Harness?

An open-source, freely composable agent harness for building and exploring agents

DeepSeek Harness developer preview (v0.1) is now open for testing by harness developers worldwide, with source code released under the MIT license. As an early preview, many details are still being refined, and the core plugins and base interfaces will iterate quickly in the weeks ahead.

DeepSeek Harness (dsh) follows an everything-is-a-plugin design: a plugin-based open architecture for building an Agent Harness — model, tools, skills, sessions, sandbox, storage, loops, scheduling, and UI are all composed from plugins, freely swappable and recomposable.

Developers can select, replace, or extend any single capability as a plugin — without ever modifying DeepSeek Harness source code itself.

Read the full introduction →

Design Philosophy

Everything Is a Plugin

Built on Cordis, a plugin system with spatiotemporal composability

Cordis

A meta-framework that does just two things

The Cordis meta-framework handles only plugin loading, unloading, and dependencies. Every concrete component of the harness is a distinct Cordis plugin.

Services & Events

Plugins cooperate with each other

Plugins cooperate through Cordis services and typed events, and can be freely composed at the configuration layer — swap one provider and you change the product.

Replaceable

Every capability can be replaced

Model adapters, tool registries, session logs, and the agent loop itself are all plugins, so each part is replaceable from configuration.

No Privileged Kernel

No kernel to patch

You extend dsh by mounting plugins alongside other plugins. Every registration is a reversible side effect, undone when its plugin unloads.

Run Modes

Multiple Run Modes

Each mode loads a different plugin set for different use cases

Standard

The full tool set

Ships the complete tool set, out of the box, for most day-to-day tasks.

PTC

Programmatic Tool Calling

The model generates a piece of code that composes multiple rounds of tool calls into a single programmatic flow.

Minimal

Benchmarking in the smallest environment

Keeps only a shell tool and a file-editing tool — for model benchmarking in a minimal environment.

Creative

Experiment and author new modes

Inspect the current runtime, try out Cordis plugins in memory, and compose new modes from them.

Observability

Every Run Is Traceable

Everything the model sees is written to an append-only session log

1

Fully recorded

System prompts, chain-of-thought, tool calls and results, sub-agent dispatch, and every context injection are written to the session log.

2

Browse by source

In the Trajectory view you can inspect this information by source, reconstructing every run's trace.

3

One event stream

Resume, fork, retrieval, and replay all share the same event stream — what the model sees is what gets recorded.

4

Rebuildable

Everything that reaches a model request must be rebuildable from the log, enforced by a runtime invariant.

Get Started

Quick Start

One command starts the Web UI

Try it via npm

On a system with the Node.js toolchain installed, start the Web UI with npx:

npx @deepseek-ai/dsh web

Run from source

Clone the full source and follow the repository instructions:

git clone https://github.com/deepseek-ai/deepseek-harness

See all install options →

FAQ

Frequently Asked Questions

Common questions about the architecture, modes, license, and community.

What is DeepSeek Harness?

DeepSeek Harness (dsh) is an open-source agent harness by DeepSeek AI. It uses an everything-is-a-plugin architecture powered by Cordis, composing model, tools, skills, sessions, sandbox, storage, loops, scheduling, and UI from swappable plugins.

What does "everything is a plugin" mean?

Every concrete component of the harness is a distinct Cordis plugin. The Cordis meta-framework handles only loading, unloading, and dependencies; plugins cooperate through services and events and are composed at the configuration layer — you can select, replace, or extend any capability without touching the source.

What run modes are available?

Four: Standard (full tool set), PTC (programmatic tool calling), Minimal (just a shell tool plus a file-editing tool for benchmarking), and Creative (inspect the runtime, try plugins in memory, and author new modes). Each loads a different plugin set.

How do I try it quickly?

Run npx @deepseek-ai/dsh web on a system with Node.js to start the Web UI (default http://127.0.0.1:3080). Or git clone https://github.com/deepseek-ai/deepseek-harness to build from source. See the install page.

Can runs be traced?

Yes. Everything the model sees is written to an append-only session log — system prompts, chain-of-thought, tool calls and results, sub-agent dispatch, and context injections. Browse by source in the Trajectory view; resume, fork, retrieval, and replay share one event stream.

Is it open source? Can I use it commercially?

Fully open source under the MIT License — use, modify, and sell freely. The repository is at github.com/deepseek-ai/deepseek-harness.

How do I join the plugin ecosystem?

Share feedback or bug reports through GitHub Discussions, and add the dsh-plugin topic to your plugin repository. See Contact.

Is this a stable release?

No — this is a developer preview (v0.1) iterating rapidly. There will be compatibility-breaking changes. Core plugins and base interfaces will keep evolving, and we welcome your feedback.

What is an agent harness?

An agent harness is the framework that runs and orchestrates agents — model, tools, sessions, sandbox, loops, and scheduling. See What Is an Agent Harness.

How does DeepSeek Harness differ from Claude Code / Codex / OpenCode?

dsh is an MIT-licensed, everything-is-a-plugin agent harness where model, tools, and sandbox are freely swappable; Claude Code and Codex are closed-source coding agents, and OpenCode is an open-source coding agent. See the comparison page.

Join Us

Explore the ceiling of intelligence,
build the DSH plugin ecosystem.

DeepSeek Harness v0.1 is only the starting point. Together with developers worldwide,
we build on open, reusable, composable infrastructure to explore what agents can do.

Join global harness developers in building the DSH plugin ecosystem.

Try It NowView Source