DeepSeek Harness v0.2 发布官方桌面应用，开源 Agent Harness 迈入开发者预览

DeepSeek 发布了官方桌面应用程序：https://x.com/DeepSeekHarness/status/2105330281389662575，用于 DeepSeek Harness：https://github.com/deepseek-ai/deepseek-harness (dsh)，这是其开源代理套件。该应用程序随 v0.2 预览版发布：https://x.com/DeepSeekHarness/status/2105330281389662575。安装程序覆盖 macOS（Apple silicon）和 Windows（64 位）。

可以部署吗？可以，从今天开始作为预览版使用。从 deepseek.com/harness 下载：https://www.deepseek.com/en/harness/ 或运行 npx @deepseek-ai/dsh web。DeepSeek 警告说将会有破坏兼容性的更改。

套件(harness) 是将模型转化为代理的运行时。它读取文件、运行命令并保留计划。v0.2 预览版面向日常工作以及编码。

桌面版本还捆绑了 dsh 命令。用户不再需要单独安装 Node.js 或 pnpm：https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.0-rc.2。用户可以使用 DeepSeek 账户登录或添加 API 密钥。根据 rc.1 注释：https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.0-rc.1，账户模型无需额外密钥即可获得网页搜索功能。

DeepSeek 于 2026 年 8 月首次将 dsh 开源，采用 MIT 许可证。它运行在 Cordis：https://github.com/cordiverse/cordis 之上，正如研究论文《一种面向时空可组合性的编程范式》https://arxiv.org/abs/2608.25512 所述。模型适配器、工具注册表和代理循环都是插件。任何一层都可以被替换。

该套件并不局限于 DeepSeek 模型。提供商指南：https://deepseek-harness.github.io/deepseek-harness/guide/providers 涵盖第三方提供商和自定义兼容 OpenAI 的端点。该仓库已获得超过 240,000 个 GitHub 星标和 29,000 个分支。

dsh 还兼作 DeepSeek 的评测工具。API 更新日志：https://api-docs.deepseek.com/updates/ 显示代码代理基准在“DeepSeek Harness 最小模式”下运行。DeepSeek-V4-Flash-0731 在 Terminal Bench 2.1 中的得分为 82.7。

根据 36Kr 的动手体验：https://eu.36kr.com/en/p/3998199345500040 和发布说明，会议可以在 4 种模式之一中运行：

v0.2.1-alpha.1 版本构建：https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.1-alpha.1 添加了实验性的 Claude Code Mods 兼容层。DeepSeek 将其视为测试 Mods API 是否广泛上是 dsh 插件的子集。它并不承诺完全兼容。该构建还添加了“让 Agent 创建插件”条目、一个可选的开发者工具包以及用于反向代理的 --public-url 标志。它移除了运行时不变插件，对某些扩展是一个重大变更。

需要与我们合作以推广您的 GitHub 仓库或 Hugging Face 页面或产品发布或网络研讨会等？请与我们联系：https://forms.gle/MJjjVDPS7whH8Ngs6

Michal Sutter 是一名数据科学专业人士，拥有帕多瓦大学数据科学硕士学位。凭借在统计分析、机器学习和数据工程方面的扎实基础，Michal 擅长将复杂的数据集转化为可操作的洞察。

Datalab 推出 OmniExtractBench，以修复提取基准中的偏差和不透明性：https://pxllnk.co/hxplrq

以实践者为中心的 AI/ML 新闻和分析，每月有超过 100 万开发者和研究人员阅读。

DeepSeek has released an official desktop app：https://x.com/DeepSeekHarness/status/2105330281389662575 for DeepSeek Harness：https://github.com/deepseek-ai/deepseek-harness (dsh), its open-source agent harness. The app ships with the v0.2 preview：https://x.com/DeepSeekHarness/status/2105330281389662575. Installers cover macOS (Apple silicon) and Windows (64-bit).

Is it deployable? Yes, today, as a preview. Download it from deepseek.com/harness：https://www.deepseek.com/en/harness/ or run npx @deepseek-ai/dsh web . DeepSeek warns that compatibility-breaking changes will follow.

A harness is the runtime that turns a model into an agent. It reads files, runs commands and keeps a plan. The v0.2 preview targets everyday work as well as coding.

The desktop build also bundles the dsh command. Users no longer need a separate Node.js or pnpm install：https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.0-rc.2. Users can sign in with a DeepSeek account or add an API key. Per the rc.1 notes：https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.0-rc.1, account models get web search without an extra key.

DeepSeek first open-sourced dsh in August 2026, under MIT. It runs on Cordis：https://github.com/cordiverse/cordis as described in the research paper ‘A Programming Paradigm for Spatiotemporal Composability：https://arxiv.org/abs/2608.25512‘. The model adapter, tool registry and agent loop are all plugins. Any layer can be replaced.

The harness is not locked to DeepSeek models. The provider guide：https://deepseek-harness.github.io/deepseek-harness/guide/providers covers third-party providers and custom OpenAI-compatible endpoints. The repository has passed 240,000 GitHub stars and 29,000 forks.

dsh also doubles as DeepSeek’s evaluation rig. The API changelog：https://api-docs.deepseek.com/updates/ says Code Agent benchmarks ran on ‘DeepSeek Harness minimal mode.’ DeepSeek-V4-Flash-0731 scored 82.7 on Terminal Bench 2.1 in that setup.

Sessions run in 1 of 4 modes, per 36Kr’s hands-on：https://eu.36kr.com/en/p/3998199345500040 and the release notes:

The v0.2.1-alpha.1 build：https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.1-alpha.1 adds an experimental Claude Code Mods compatibility layer. DeepSeek frames it as a test of whether the Mods API is broadly a subset of dsh plugins. It does not promise full compatibility. The build also adds a ‘Let Agent create a plugin’ entry, an optional Developer Tools bundle and a --public-url flag for reverse proxies. It removes runtime invariant plugins, a breaking change for some extensions.

Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us ：https://forms.gle/MJjjVDPS7whH8Ngs6

Michal Sutter is a data science professional with a Master of Science in Data Science from the University of Padova. With a solid foundation in statistical analysis, machine learning, and data engineering, Michal excels at transforming complex datasets into actionable insights.

Datalab Introduces OmniExtractBench to Fix Bias and Opacity in Extraction Benchmarks ：https://pxllnk.co/hxplrq

Practitioner-first AI/ML news and analysis, read by 1M+ developers and researchers every month.