DeepSeek Harness is now an official open-source agent framework from DeepSeek. Released as a developer preview, it provides a plugin-first runtime for models, tools, skills, sessions, sandboxes, storage, orchestration, and a web interface.
Independent source-based guide. DeepSeek Harness is available under the MIT License as a developer preview; its APIs and plugin contracts may still change before a stable release.
Released
Aug 13, 2026
DeepSeek published the official Harness website, source repository, and developer preview package.
License
MIT
The framework is open source and can be inspected, modified, and self-hosted under the MIT License.
Status
Developer Preview
The first public release is usable today, but DeepSeek explicitly warns that compatibility may change.
Community plugin directory
Extend the harness with one focused capability
Browse 316 open-source projects collected from the GitHub dsh-plugin topic. Switch categories, search by name, and open any plugin for its source details.
316
Source snapshots indexed from GitHub and organized into eight practical categories.
7d7d —— 7k7k 风格的 DSH 游戏门户：在 Web UI 内生成、同步并游玩 HTML5 与自托管 Ruffle Flash 小游戏。
omdsh-dev/7d7d
Portable agent skills for building , publishing , and remixing games on Paean Apps Square ( .clide.app / 8x.gg ) — usable from Zero CLI , Claude Code , and Codex (or any agent that can read a SKILL.md and run a Node script).
paean-ai/8x-skills
Abu - AI Desktop Office Assistant
PM-Shawn/Abu-Cowork
Media & Vision
agent-vision-toolkit
What it thinks is what it sees — give any text-only coding agent eyes: image Q&A, long-screenshot OCR, frontend UI restoration, and GUI automation, as a vision toolkit plus a skill, with optional drop-in integration for Codex, Claude Code, Pi, Oh My Pi, and OpenCode.
Anionex/agent-vision-toolkit
Development
ai-sdk-provider-dsh
AI SDK provider that drives a DeepSeek Harness (dsh) runtime as a language model (LanguageModelV3 — works on both AI SDK v6 and v7)
krislavten/ai-sdk-provider-dsh
"Any claim must have a verifiable practice anchor." — First Law, Materialist Practice Theory
unknowbug/anchorlaw
Argo 阿尔戈 — 统一搜索与证据核验。推荐 npx -y github:taxueseek/argo 或 install.sh 安装（不依赖 npm 发版）。多语言路由、120+ 引擎、影视/体育/金融等垂直源、垂直结构化模态卡、10 个 MCP 工具。
taxueseek/argo
Utilities
awesome-deepseek-harness
A curated DeepSeek Harness ecosystem directory covering plugins, tools, and infrastructure.
0xsline/awesome-deepseek-harness
Integrations
awesome-deepseek-harness
A curated list of plugins, skills, MCP servers, orchestrators, aggregators & UIs for DeepSeek Harness (DSH) — DeepSeek's official agent runtime built around the idea Model + Harness = Agent .
Dominic789654/awesome-deepseek-harness
Integrations
awesome-dsh-bridges
Bridge your favorite AI coding tools into DeepSeek Harness -- zero migration, full compatibility.
YYTbit/awesome-dsh-bridges
Media & Vision
awesome-DSH-plugin
A community-curated, vendor-neutral catalog of plugins for DeepSeek Harness (DSH) — from developer tooling and data workflows to media, operations, and everyday life.
Alex-Yanggg/awesome-DSH-plugin
Utilities
awesome-dsh-plugin
A curated list of plugins for DeepSeek Harness ( dsh ).
awesome-dsh-plugin/awesome-dsh-plugin
What DeepSeek Harness means right now
The earlier hiring signal has become a public release. DeepSeek now describes Harness as an open-source framework for building and running AI agents, with a plugin architecture, traceable sessions, multiple runtime modes, and a browser-based interface.
Release channel
Developer Preview
The official site and repository present this as an early public release rather than a stable production milestone.
Install path
npx or source
Run the published package to launch the web interface, or clone the MIT-licensed repository for development.
Architecture
Plugin-first
Models, tools, skills, sessions, storage, sandboxes, loops, scheduling, and UI are exposed through plugins.
The verified release
The strongest evidence is now first-party: DeepSeek's product page, official announcement, public repository, package, and documentation.
Model + Harness = Agent
DeepSeek's release turns the earlier product thesis into software: the model remains one component, while the harness supplies the runtime, tools, state, and interface around it.
Official developer preview
DeepSeek now publishes Harness under its own name, with an official website, installation command, documentation, and public source code.
MIT-licensed source
The repository makes the implementation inspectable and extendable, while the preview label keeps expectations appropriately cautious.
Everything is a plugin
Built on Cordis, the framework exposes its major runtime capabilities as plugins instead of presenting a fixed, closed coding assistant.
Why the harness matters more than another coding demo
A code harness is the product and runtime layer around a model. It decides how the model sees a repository, chooses tools, plans work, runs commands, remembers state, handles failures, and turns real task traces into product and model improvements.
Repository context
Indexes files, instructions, diffs, dependencies, tests, and prior decisions so the model works from the actual project state instead of a thin prompt.
Tool execution
Connects the agent to shell commands, editors, search, browsers, test runners, package managers, and external services with clear boundaries.
Runtime modes and orchestration
Offers Standard, Code, Minimal, and Creator modes, with loop, scheduling, and orchestration plugins available for different agent workflows.
Traceable session state
Stores sessions as append-only event logs so the context shown to a model can be traced, resumed, forked, searched, and replayed.
How a coding agent harness works
The released architecture follows a recognizable agent loop while making the surrounding runtime modular. This is a practical reading of the public framework, not a claim that every plugin behaves identically.
Why this release matters
DeepSeek already has strong developer mindshare around model capability and cost. A first-party, open-source harness now gives it a direct role in the developer workflow around those models.
AI coding is becoming agentic
The market is shifting from autocomplete to tools that plan, modify projects, run tests, and complete multi-step engineering tasks.
Long agent loops reward efficient models
Coding agents can spend many turns reading, editing, and testing. Lower inference cost can become a practical advantage if the harness is reliable.
Owning the harness owns the workflow
The interface, trace, permission model, integration surface, and recovery experience shape developer loyalty as much as raw benchmark scores.
What to watch next
Preview stability
DeepSeek labels the release a developer preview and warns that APIs and plugin contracts may change as the framework evolves.
Interface coverage
The current entry point emphasizes a local web interface. The package also exposes a CLI, but the project is not yet presented as a mature terminal-first or IDE-native product.
Plugin ecosystem
The architecture is intentionally extensible; practical value will depend on plugin quality, provider support, compatibility, and the pace of community maintenance.
Sources used for this guide
DeepSeek Harness is still early enough that source discipline matters. This page now prioritizes official release materials while retaining the original hiring signal as historical context.
Questions people are asking about DeepSeek Harness
The important line is now between what the developer preview ships today and what may change before a stable release. These answers keep that distinction explicit.
From hiring signal to public preview
The original role listings document how the Harness team formed. The framework is now public through DeepSeek's official site, package, and GitHub repository; these listings remain useful historical sources.