dsh-web-search

Host and Web client plugin for selecting and routing DeepSeek Harness web_search backends.

---

English

What it does

The Host half registers one ctx.web provider named dsh-web-search-selected. A persisted web-search setting chooses the live backend. The Web client adds a dedicated Settings → Web search section and saves the selection through the Host settings scope.

Request path:

web_search
  -> ctx.web
  -> dsh-web-search-selected
  -> selected backend

Built-in backends:

ID	Behavior
openai-responses	OpenAI Responses API with the native web_search tool
deepseek-official	DeepSeek native web search protocol

The plugin identity is the router, not any one model or vendor.

Install

dsh plugin --profile web add github:RailgunHamster/dsh-web-search
dsh plugin --profile web install

Restart the Web profile:

dsh web

Or add it to the Web profile package.json by hand:

{
  "dependencies": {
    "dsh-web-search": "git+https://github.com/RailgunHamster/dsh-web-search.git"
  },
  "dsh": {
    "profile": {
      "bundles": [
        "dsh-web-search"
      ]
    }
  }
}

cd "$env:USERPROFILE\.dsh\profiles\web"
pnpm install
dsh web

The bundled cordis.patch.yml then:

1. Disables the shipped web-search-deepseek provider.
2. Points @deepseek-ai/dsh-web searchProvider at dsh-web-search-selected.
3. Inserts dsh-web-search into the current composition.

Configuration

Plugin config (and the settings namespace web-search) accepts:

Key	Default	Meaning
provider	openai-responses	Selected backend
responsesApiKeyEnv	OPENAI_API_KEY	Credential / env name for Responses search
responsesBaseURL	https://api.openai.com/v1	Responses API base URL
responsesModel	gpt-4o	Model that exposes tools: [{ type: "web_search" }]
responsesMaxTokens	1024	max_output_tokens for Responses search
deepseekApiKeyEnv	DEEPSEEK_API_KEY	Credential / env name for DeepSeek search
deepseekBaseURL	https://api.deepseek.com/anthropic/v1	DeepSeek Anthropic-compatible base URL
deepseekModel	deepseek-v4-flash	DeepSeek search model
deepseekApiVersion	2023-06-01	anthropic-version header
deepseekMaxTokens	4096	DeepSeek max_tokens
deepseekMaxUses	5	DeepSeek web_search max_uses

Provide keys through DSH credentials or the launching environment. Do not put real keys in Git, settings.yaml, or this README.

Example plugin row:

- id: web-search
  config:
    provider: openai-responses
    responsesApiKeyEnv: OPENAI_API_KEY
    responsesBaseURL: https://api.openai.com/v1
    responsesModel: gpt-4o

Switching backends in Settings → Web search saves immediately and applies to later web_search calls. A DSH restart is not required for that switch.

Verify

dsh --profile web --dump-config | Select-String -Pattern 'searchProvider|web-search-deepseek|web-search'

Expected:

searchProvider: dsh-web-search-selected
web-search-deepseek: disabled: true
web-search: dsh-web-search

Then run a normal web_search. The default Responses backend posts:

POST {responsesBaseURL}/responses
tools: [{ "type": "web_search" }]

The plugin maps returned text and url_citation annotations to Harness content and sources.

Development

Host or cordis.patch.yml changes need a full DSH Web restart, then a browser refresh. Client changes hot-reload only if the profile Web watcher is running.

Do not commit:

• node_modules/
• .dsh/
• settings.yaml
• credential files
• .env files
• API keys, private keys, or session data

License

MIT

---

中文

功能

Host 侧注册一个名为 dsh-web-search-selected 的 ctx.web provider。持久化的 web-search 设置决定当前搜索后端。Web 客户端提供独立的 设置 → 联网搜索 页面，选择会通过 Host settings scope 自动保存。

请求路径：

web_search
  -> ctx.web
  -> dsh-web-search-selected
  -> 当前选中的 backend

内置 backend：

ID	行为
openai-responses	调用 OpenAI Responses API 的原生 web_search 工具
deepseek-official	使用 DeepSeek 原生 web search 协议

插件本身是路由器，不绑定某个模型或厂商。

安装

dsh plugin --profile web add github:RailgunHamster/dsh-web-search
dsh plugin --profile web install

然后重启 Web profile：

dsh web

也可以手动改 Web profile 的 package.json：

{
  "dependencies": {
    "dsh-web-search": "git+https://github.com/RailgunHamster/dsh-web-search.git"
  },
  "dsh": {
    "profile": {
      "bundles": [
        "dsh-web-search"
      ]
    }
  }
}

cd "$env:USERPROFILE\.dsh\profiles\web"
pnpm install
dsh web

插件自带的 cordis.patch.yml 会：

1. 禁用 shipped 的 web-search-deepseek provider。
2. 将 @deepseek-ai/dsh-web 的 searchProvider 指向 dsh-web-search-selected。
3. 把 dsh-web-search 插入当前 composition。

配置

插件配置（以及 settings namespace web-search）支持：

字段	默认值	含义
provider	openai-responses	当前 backend
responsesApiKeyEnv	OPENAI_API_KEY	Responses 搜索使用的凭据 / 环境变量名
responsesBaseURL	https://api.openai.com/v1	Responses API 根地址
responsesModel	gpt-4o	支持 tools: [{ type: "web_search" }] 的模型
responsesMaxTokens	1024	Responses 搜索的 max_output_tokens
deepseekApiKeyEnv	DEEPSEEK_API_KEY	DeepSeek 搜索使用的凭据 / 环境变量名
deepseekBaseURL	https://api.deepseek.com/anthropic/v1	DeepSeek Anthropic 兼容根地址
deepseekModel	deepseek-v4-flash	DeepSeek 搜索模型
deepseekApiVersion	2023-06-01	anthropic-version 请求头
deepseekMaxTokens	4096	DeepSeek max_tokens
deepseekMaxUses	5	DeepSeek web_search 的 max_uses

密钥只能放在 DSH credentials 或启动环境里，不要写入 Git、settings.yaml 或本文。

插件行示例：

- id: web-search
  config:
    provider: openai-responses
    responsesApiKeyEnv: OPENAI_API_KEY
    responsesBaseURL: https://api.openai.com/v1
    responsesModel: gpt-4o

在 设置 → 联网搜索 里切换 backend 会立刻保存，并作用于之后的 web_search，不必为了切换而重启 DSH。

验证

dsh --profile web --dump-config | Select-String -Pattern 'searchProvider|web-search-deepseek|web-search'

预期看到：

searchProvider: dsh-web-search-selected
web-search-deepseek: disabled: true
web-search: dsh-web-search

然后发一次普通 web_search。默认 Responses backend 会请求：

POST {responsesBaseURL}/responses
tools: [{ "type": "web_search" }]

插件把返回文本和 url_citation 映射为 Harness 的 content 与 sources。

开发

Host 代码或 cordis.patch.yml 改动后需要完整重启 DSH Web，再刷新浏览器。只有 profile 的 Web watcher 在跑时，Client 改动才会热更新。

不要提交：

• node_modules/
• .dsh/
• settings.yaml
• credentials 文件
• .env 文件
• API key、私钥或 session 数据

许可证

MIT