hiyms/dsh-web-search-provider

改进Deepseek Harness中的网络搜索,使用服务器端搜索服务。

Project Overview项目介绍

This is a native web search plugin built exclusively for DeepSeek Harness (DSH), that leverages the model provider's built-in server-side web search capability instead of DSH's native tool calling flow. It can be installed directly from the DSH plugin registry with a single command, or built from source and added to your cordis.yml configuration file manually. Because it does not conflict with DSH's built-in web search tool, you can run it alongside other existing search plugins like the default web-search-deepseek without issues. To use it, you need a model provider that supports either the OpenAI Responses API or an Anthropic-compatible Messages API, and enables server-side web search for its models.

Independent tests have shown that this plugin outperforms DSH's built-in web-search-deepseek in both search speed and token consumption. In tests running 5 searches 10 times on the same Deepseek V4 Flash model with high thinking capability, this plugin averaged 822 output tokens and 14.5 seconds per run, compared to 4446 tokens and 47.5 seconds for the built-in plugin. The advantage grows larger for tasks that require multiple web searches or reading content directly from web pages, especially when using a provider that supports the OpenAI Responses API. It is ideal for any DSH user that regularly performs web research for work or personal projects.

This plugin is released under the permissive MIT open source license, so it is free to use and modify, but note that some model providers charge extra fees for enabling server-side web search. All configuration options are optional, and you can adjust settings like the enable toggle, provider whitelist, custom endpoint URL, model override, and timeout limits. Known limitations include that it will throw an error on first use if your provider does not support server-side search, and there is no visible UI for search calls. Errors do not break DSH core functionality, so the app will continue working normally even if the plugin fails.

这是一个专为DeepSeek Harness开发的原生网络搜索插件,利用模型提供商自带的服务端搜索能力实现网络搜索,而非依赖DSH内置的工具调用流程,因此可以和其他网络搜索插件(如内置的web-search-deepseek)并行使用。使用该插件要求模型提供商支持OpenAI Responses API或兼容Anthropic的Messages API,且本身开放网络搜索能力,目前已验证可配合DeepSeek官方API、OpenCode Go的两种API以及OpenAI转发站API使用。

安装方式非常简便,只需要在DSH的命令行界面执行dsh plugin add dsh-web-search-provider即可完成安装,也可以拉取源码后通过pnpm构建,再在cordis.yml配置文件中添加插件条目启用。相比DSH内置的web-search-deepseek插件,它的单次搜索速度更快,消耗的输出token更少,在需要大量网络搜索的任务中优势更加明显,适合经常需要借助网络搜索获取最新信息的DSH用户使用。

本插件采用MIT许可证开源,无使用成本,但部分模型提供商可能对服务端搜索能力额外收费。配置项都为可选,可设置启用开关、提供商白名单、自定义端点地址、模型覆盖、超时时间等参数。已知限制包括:不支持无服务端搜索能力的模型提供商,且使用搜索时没有可见的搜索调用UI,出错后不会影响DSH本身正常运行。

Pre-install check安装前体检Compatibility · Security兼容性 · 安全性 1 warning1 项注意
  • Only 4 stars - very few users, little community feedback星标只有 4,几乎没人在用,遇到问题缺少社区反馈
DSH walks through these 9 checksDSH 会逐条核对这 9 项

Compatibility兼容性

  • DSH, Node, OS and profile requirementsDSH 版本 / Node 版本 / 操作系统 / profile 是否满足要求
  • External dependencies and runtimes (Electron / Python / Docker, ...)外部依赖与运行时(Electron / Python / Docker 等)是否齐备
  • Conflicts with installed plugins: command names, skill / tool names, ports, duplicate MCP registration与已装插件是否冲突:命令名、skill / tool 重名、端口占用、重复 MCP 注册

Security安全性

  • Repo matches the facts registered here; archived or abandoned?仓库是否与页面登记一致,是否归档或长期停更
  • Safety of preinstall / install / postinstall and install.sh / setup.ps1preinstall / install / postinstall 与 install.sh、setup.ps1 是否安全
  • curl|bash, download-then-execute, obfuscation, unrelated domains → stop immediatelycurl|bash、下载即执行、混淆代码、无关域名 → 立刻停止
  • Typosquatting or unmaintained packages among the new dependencies新增依赖里有没有 typosquatting 或无人维护的包
  • Requested permissions vs. what the feature actually needs申请了哪些权限、是否超出功能所需(filesystem / network / shell / clipboard)
  • Any sudo / admin requirement, plus uninstall and rollback是否要求 sudo / 管理员权限,以及卸载与回滚方式

Anything uncertain must be marked unknown with a note on how to confirm it. This site's signal screen is a static snapshot, not a security audit.拿不准的必须标「未知」并说明要我怎么确认。本站的信号筛查是静态快照,不能替代安全审计。

Or use CLI install (for developers)或使用命令行安装(适合开发者)

CLI Install命令行安装

dsh plugin add dsh-web-search-provider

把 hiyms/dsh-web-search-provider 加入你的 DSH 配置(web profile)即可启用。

READMEREADME

dsh-web-search-provider

npm version

简体中文

Network search support for the Deepseek Harness, powered by the model provider's server-side capability.

Using this plugin requires the model provider to use the OpenAI Responses API or an Anthropic-compatible Messages API, and to provide a web search capability.

[!TIP] The Anthropic-compatible Messages API only provides a search capability, while the OpenAI Responses API additionally supports viewing a URL's page content / finding specific content within a URL's page. Providers on the OpenAI Responses API are recommended.

[!TIP] Verified to work with: the Deepseek official Messages API (the DSH built-in Deepseek provider), the OpenCode Go Messages API, the OpenCode Go Response API, and an OpenAI relay station's Response API.

[!WARNING] The DSH built-in OpenCode Go provider uses different API types for different models. For example, for its Deepseek V4 Flash/Pro models, to use this plugin's capability you need to manually add a custom provider of the Response / Message API type.

[!WARNING] Some model providers may charge extra for the web search capability this plugin uses.

This plugin does not provide its support through the web_search tool inside the Deepseek Harness, so you can use it alongside other web search provider plugins (such as the built-in web-search-deepseek).

Comparing with web-search-deepseek

TL;DR Compared with web-search-deepseek, this plugin has a clear advantage in both token consumption and time.

web-search-deepseek is the Deepseek Harness's built-in search plugin: when the model in a session calls the web_search tool, it creates a new internal session to run the search and returns several URLs with their content summaries.

Compared with web-search-deepseek, the AI in this plugin's sessions sends the network search request directly to the server, making each individual search faster. Also, the model can directly view / search page content (only with the OpenAI Responses API), reducing the tokens the AI spends on bash/curl to read a page's full content.

Showing the opening section of the README — the full document lives in the repository以上为 README 开头摘要,完整文档在仓库内 · View the full README on GitHub →在 GitHub 查看完整 README →

← 上一个 Prev dsh-launcher 下一个 Next dsh-custom-ui →