dsh-free-search 0.4.38 → 0.5.0

package/README.md

@@ -29,7 +29,7 @@ dsh 默认的搜索 provider 依赖 DeepSeek 官方 API key（`DEEPSEEK_API_KEY`
 ### 特性
 
 - **零成本** —— 多个免费引擎，无需 key、无需注册
-- **多引擎可选**：DuckDuckGo（html/lite）、Bing、SearXNG（元搜索，支持自定义实例）、AnySearch、Exa、Tavily、Keenable、Firecrawl、Parallel、Perplexity、SerpBase、DeepSeek 官方
+- **多引擎可选**：DuckDuckGo（html/lite）、Bing、SearXNG（元搜索，支持自定义实例）、AnySearch、Exa、Tavily、Keenable、Firecrawl、Parallel、Perplexity、SerpBase、DeepSeek 官方、You.com
 - **网页设置页** —— 引擎切换 + API key 配置（UI 中 key 脱敏显示"已配置"）+ 中英文切换；入口在左侧「插件」页的组件行配置（`plugins.row.config`，DSH 0.1.7-rc.1+）
 - **弹出式切换命令** —— 聊天框输入 `/free-search-engine`，弹出引擎选择窗口，点选即切换（等效设置页 + 保存）
 - **引擎测试** —— `free_search_test` 工具让 agent 一键测试所有引擎；设置页也有"测试引擎"按钮（直测当前引擎，不走回退链，付费引擎无 key 会明确报错）
@@ -50,6 +50,7 @@ dsh 默认的搜索 provider 依赖 DeepSeek 官方 API key（`DEEPSEEK_API_KEY`
 
 | id | 引擎 | 费用 | 说明 |
 |---|---|---|---|
+| `auto` | Auto 智能路由 | 动态 | **根据查询语言/时间条件自动选路**（含中文优先 Bing/Baidu/Aliyun/AnySearch，英文优先 Bing/Exa/Tavily；时间过滤优先支持引擎），末尾全量回退 |
 | `ddg` | DuckDuckGo HTML | 免费 | 偶发限流（反爬），解封自动恢复 |
 | `ddg-lite` | DuckDuckGo Lite | 免费 | 轻量版，同上 |
 | `bing` | Bing | 免费 | **默认引擎**，最稳定，中文优化（zh-CN） |
@@ -63,6 +64,10 @@ dsh 默认的搜索 provider 依赖 DeepSeek 官方 API key（`DEEPSEEK_API_KEY`
 | `perplexity` | Perplexity | 付费 | 需 `PERPLEXITY_API_KEY` |
 | `serpbase` | SerpBase | 付费 | 需 `SERPBASE_API_KEY`（serpbase.dev，注册送 100 次免费额度） |
 | `deepseek-official` | DeepSeek 官方 | 付费 | 需 `DEEPSEEK_API_KEY` |
+| `you` | You.com | 付费 | 需 `YOUCOM_API_KEY`（you.com/platform/api-keys，注册即送免费额度） |
+| `baidu` | 百度千帆 AI 搜索 | 额度 | 需 `BAIDU_API_KEY`（千帆 AI 搜索，每日赠 50 次，超额按量后付费）；中文全网，支持时间过滤 |
+| `kimi` | Kimi（Moonshot）联网搜索 | 付费 | 需 `MOONSHOT_API_KEY`（basic 约 ￥0.01/次），返回带正文 chunks 的中文结果 |
+| `aliyun` | 阿里云百炼 EnhancedSearch | 付费 | 需 `DASHSCOPE_API_KEY`（MCP search_pro，约 ￥0.03/次，新用户 200 次免费包）；中文全网带来源 hostname |
 
 - **默认引擎为 `bing`**（免费且最稳定），安装后开箱即用。
 - **自动回退**：任何引擎失败（免费限流/反爬，付费缺 key/无效/网络错误）都会自动轮流尝试下一个引擎——先试其他已配 key 的付费引擎，再试免费引擎（Bing/AnySearch 等），并在结果中附带回退提示——搜索不会因引擎问题直接失败。
@@ -74,6 +79,7 @@ dsh 默认的搜索 provider 依赖 DeepSeek 官方 API key（`DEEPSEEK_API_KEY`
   - Perplexity：<https://www.perplexity.ai/settings/api>
   - SerpBase：<https://serpbase.dev>
   - DeepSeek：<https://platform.deepseek.com/api_keys>
+  - You.com：<https://you.com/platform/api-keys>
 
 #### 为什么免费引擎不需要 key？
 
@@ -96,6 +102,31 @@ dsh plugin --profile web add /path/to/dsh-free-search
 dsh web
 ```
 
+#### 接管行为与验证
+
+插件加载后**自动接管搜索**：
+
+- `web.searchProvider` 未设置，或仍是 DSH 出厂默认的官方搜索 `deepseek-official` 时，自动切换为本插件；
+- 如果（你或别的插件）已显式选择其他 provider，本插件不抢占，只在启动日志输出 WARN 与切换用的 YAML。
+
+正常安装（`dsh plugin add` / 插件管理器）时，插件自带的 **bundle patch**（`cordis.patch.yml`）还会在配置层显式写入 `searchProvider: ddg`；即使这层没生效（例如把插件作为普通依赖手工安装、或 profile patch 整体覆盖了 `web` 条目），上面的运行时兜底也会接管官方默认，不会再静默退回官方搜索。
+
+需要显式声明、或从其他 provider 切过来时（`profiles/<profile>/cordis.patch.yml`）：
+
+```yaml
+# 让 harness 的 web_search 使用本插件。
+# DSH 0.1.2+ 的 patch 是"整段覆盖 config"而不是深度合并：
+# 已有 web 条目里的字段（如 fetchProvider）必须在这里一并重述，否则会被抹掉。
+- id: web
+  config:
+    searchProvider: ddg
+    fetchProvider: http
+```
+
+- `searchProvider: ddg` —— `ddg` 是**本插件注册的 provider id（固定值）**，不是"使用 DuckDuckGo 引擎"的意思；具体用哪个引擎由设置页的 `provider` 字段决定（可填 `bing`/`baidu`/`auto` 等）。
+- `fetchProvider: http` —— 官方网页抓取（web-fetch-http），请保留；漏掉会导致网页抓取失效或重复注册。
+- 想改回官方搜索：在插件管理器里停用本插件条目即可。
+
 #### 姊妹插件：dsh-preset-workbench（预设工作台）
 
 同作者的**姊妹插件**：在设置页里可视化创建/编辑 Agent 预设——分段提示词、15 项能力开关、内置「鲸鱼娘 / 梁神模式」模板，不用手写 YAML。两者搭配：**free-search 解决"AI 联网搜索"、preset-workbench 解决"AI 人设能力编排"**，都是纯免费、开箱即用。
@@ -121,7 +152,7 @@ dsh web
 配置页提供：
 
 - **Search engine**：下拉框切换引擎，保存即生效
-- **API keys**：为 Exa / Tavily / Keenable / Firecrawl / Parallel / Perplexity / DeepSeek 填写 key（密码框，保存后只显示"已配置"；Exa / Tavily / Keenable / Firecrawl / Parallel 不填也可免 key 使用）
+- **API keys**：为 Exa / Tavily / Keenable / Firecrawl / Parallel / Perplexity / DeepSeek / SerpBase / You.com 填写 key（密码框，保存后只显示"已配置"；Exa / Tavily / Keenable / Firecrawl / Parallel 不填也可免 key 使用）
   - **推荐**：付费引擎 key 建议写入 harness 凭据中心 `~/.dsh/.credentials.yaml`（如 `DEEPSEEK_API_KEY: sk-...`，与官方 LLM provider 一致，一处管理所有 key）。插件读取优先级：凭据中心 > 设置页 > 环境变量，设置页填的 key 仅作为遗留兼容。
 - **Test engine**：直测当前引擎可用性（不走回退链，付费引擎无 key 会明确报错）
 - **Use Bing default**：把当前搜索引擎切回稳定的免费 Bing；`Discard` 只撤销尚未保存的编辑
@@ -159,7 +190,7 @@ DSH 0.1.7-rc.1 起，配置跟随 profile 的插件条目保存：设置页与 `
 
 ```yaml
 # profiles/<profile>/cordis.patch.yml 中该条目的 config：
-provider: bing              # ddg / ddg-lite / bing / searxng / anysearch / exa / tavily / keenable / firecrawl / parallel / perplexity / serpbase / deepseek-official
+provider: bing              # ddg / ddg-lite / bing / searxng / anysearch / exa / tavily / keenable / firecrawl / parallel / perplexity / serpbase / deepseek-official / you
 lang: zh                    # 设置页界面语言（zh / en）
 bingMarket: zh-CN           # Bing 市场
 region: cn-zh               # DuckDuckGo 区域（可选）
@@ -217,6 +248,15 @@ Search engine test:
 
 示例对话：*"帮我搜最近 3 天关于 DSH 的新闻"* → agent 调用 `advanced_search`，`timeRange: "3d"`。
 
+#### 多源并发合并搜索（multi_search）
+
+当需要对重要问题做**多源交叉验证**、避免单一引擎偏差或单源死锁时，可以让 agent 调用 `multi_search` 工具：
+
+- **并发请求**：默认基于当前查询类型并发请求前 3 个优选引擎（或显式传入 `engines` 列表），各引擎独立解析 API Key 与容错（缺 key 引擎自动跳过，不阻断其他引擎）。
+- **去重与合并**：按规范化 URL 去除结尾斜杠并合并结果，多引擎共同命中的条目优先置顶排在最前，并在结果附带 `seenIn` 命中来源清单（如 `[seen in: bing, exa]`）。
+- **不可信边界与清洗**：严格遵守 `<untrusted-web-content>` 数据边界，正文 snippet 统一清洗。
+- ⚠️ 注：多源并发会消耗更多 API 配额，建议在需要多角度核验时按需使用。
+
 #### 抓取网页内容（web_fetch）
 
 搜索到 URL 后，可以让 agent **读取网页全文**（如"打开第一个链接看看内容"）。`web_fetch` 工具已启用（官方 `dsh-web-fetch-http` provider）：
@@ -298,7 +338,7 @@ This plugin provides multiple free search engines with automatic fallback, compl
 ### Features
 
 - **Zero Cost** — Multiple free engines with no API key or registration required
-- **Multi-Engine Support** — DuckDuckGo (HTML / Lite), Bing, AnySearch AI, SearXNG (meta-search with custom instances), Exa, Tavily, Keenable, Firecrawl, Parallel, Perplexity, SerpBase, and DeepSeek Official
+- **Multi-Engine Support** — DuckDuckGo (HTML / Lite), Bing, AnySearch AI, SearXNG (meta-search with custom instances), Exa, Tavily, Keenable, Firecrawl, Parallel, Perplexity, SerpBase, DeepSeek Official, and You.com
 - **Web Settings UI** — Engine switching, API key configuration (keys masked as "configured" in the UI), and a Chinese/English toggle; the entry is the component-row config (`plugins.row.config`) on the sidebar Plugins page (DSH 0.1.7-rc.1+)
 - **Popup Switch Command** — Type `/free-search-engine` in the chat: a picker opens with all engines; click one to switch (equivalent to the settings page + save)
 - **Engine Testing** — `free_search_test` for the agent to check all engines in one call; the settings UI also has a "Test engine" button that tests the selected engine directly (no fallback chain; paid engines without a key report an explicit error)
@@ -319,6 +359,7 @@ If this plugin has been helpful, a ⭐ on [GitHub](https://github.com/DDDMUC/dsh
 
 | id | Engine | Cost | Description |
 |---|---|---|---|
+| `auto` | Auto Smart Routing | Dynamic | **Smartly routes engines based on query language/time filter** (Chinese queries prioritize Bing/Baidu/Aliyun/AnySearch, English queries prioritize Bing/Exa/Tavily; time filters prioritize time-capable engines), with full fallback |
 | `ddg` | DuckDuckGo HTML | Free | Occasional rate limits (anti-bot challenges); recovers automatically |
 | `ddg-lite` | DuckDuckGo Lite | Free | Lightweight version; same rate-limit behavior as above |
 | `bing` | Bing | Free | **Default engine**, most stable, optimized for Chinese (`zh-CN`) |
@@ -332,6 +373,10 @@ If this plugin has been helpful, a ⭐ on [GitHub](https://github.com/DDDMUC/dsh
 | `perplexity` | Perplexity | Paid | Requires `PERPLEXITY_API_KEY` |
 | `serpbase` | SerpBase | Paid | Requires `SERPBASE_API_KEY` (serpbase.dev, 100 free queries on signup) |
 | `deepseek-official` | DeepSeek Official | Paid | Requires `DEEPSEEK_API_KEY` |
+| `you` | You.com | Paid | Requires `YOUCOM_API_KEY` (you.com/platform/api-keys, free tier on signup) |
+| `baidu` | Baidu Qianfan AI Search | Quota | Requires `BAIDU_API_KEY` (Qianfan AI Search, 50 free calls/day then pay-as-you-go); Chinese web-wide, supports time filtering |
+| `kimi` | Kimi (Moonshot) Web Search | Paid | Requires `MOONSHOT_API_KEY` (basic ~¥0.01/call); returns Chinese results with body chunks |
+| `aliyun` | Aliyun Bailian EnhancedSearch | Paid | Requires `DASHSCOPE_API_KEY` (MCP search_pro, ~¥0.03/call, 200 free calls for new users); Chinese web-wide with source hostnames |
 
 - **Default engine is `bing`** (free and most stable), ready to use out of the box after installation.
 - **Auto-failover**: any engine failure (rate-limited free engine, or missing/invalid paid key, network error) automatically tries the next engine — the configured engine first, then other engines (exa/tavily/keenable/firecrawl/parallel are tried even without a key because they have built-in keyless quota), then the remaining free engines (Bing/AnySearch etc.) — with a note attached to the results naming the engine that actually served them (e.g. `Note: perplexity unavailable or failed, using exa.`). Search never fails outright because of engine issues.
@@ -343,6 +388,7 @@ If this plugin has been helpful, a ⭐ on [GitHub](https://github.com/DDDMUC/dsh
   - Perplexity: <https://www.perplexity.ai/settings/api>
   - SerpBase: <https://serpbase.dev>
   - DeepSeek: <https://platform.deepseek.com/api_keys>
+  - You.com: <https://you.com/platform/api-keys>
 
 #### Why are some engines free?
 
@@ -365,6 +411,31 @@ Then restart:
 dsh web
 ```
 
+#### Takeover behavior and verification
+
+The plugin **takes over search automatically** once loaded:
+
+- When `web.searchProvider` is unset, or still the shipped default `deepseek-official`, it switches to this plugin;
+- If you (or another plugin) explicitly selected a different provider, it does not steal it — it only logs a WARN with a copy-pasteable YAML snippet.
+
+With a normal install (`dsh plugin add` / plugin manager) the bundled **bundle patch** (`cordis.patch.yml`) additionally writes `searchProvider: ddg` into the config layer. Even if that layer does not apply (e.g. the plugin was added as a plain dependency, or a profile patch replaced the whole `web` entry), the runtime fallback above still takes over from the official default — search no longer silently falls back to the official provider.
+
+To declare it explicitly, or to switch over from another provider (`profiles/<profile>/cordis.patch.yml`):
+
+```yaml
+# Route the harness web_search through this plugin.
+# Since DSH 0.1.2 a patch REPLACES the whole entry config (no deep merge):
+# restate any existing web.* fields (e.g. fetchProvider) or they are dropped.
+- id: web
+  config:
+    searchProvider: ddg
+    fetchProvider: http
+```
+
+- `searchProvider: ddg` — `ddg` is this plugin's **fixed provider id**, not "use the DuckDuckGo engine". The engine is chosen by the `provider` field in the settings page (`bing` / `baidu` / `auto` / …).
+- `fetchProvider: http` — the official web-fetch provider; keep it, or page fetching breaks or double-registers.
+- To go back to official search: disable this plugin's entry in the plugin manager.
+
 #### Sister Plugin: dsh-preset-workbench
 
 A **sister plugin** by the same author: a **visual workbench for creating/editing agent presets** right inside Settings — sectioned prompts, 15 capability toggles, and built-in "Whale Girl / Liangshen Mode" templates, no YAML needed. Pair them up: **free-search gives your AI web search, preset-workbench shapes its persona & capabilities** — both free and zero-config.
@@ -428,7 +499,7 @@ Since DSH 0.1.7-rc.1 the configuration is stored with the profile's plugin entry
 
 ```yaml
 # config of that entry in profiles/<profile>/cordis.patch.yml:
-provider: bing              # ddg / ddg-lite / bing / searxng / anysearch / exa / tavily / keenable / firecrawl / parallel / perplexity / serpbase / deepseek-official
+provider: bing              # ddg / ddg-lite / bing / searxng / anysearch / exa / tavily / keenable / firecrawl / parallel / perplexity / serpbase / deepseek-official / you
 lang: zh                    # settings UI language (zh / en)
 bingMarket: zh-CN           # Bing market
 region: cn-zh               # DuckDuckGo region (optional)
@@ -486,6 +557,15 @@ Ask the agent for *"news from the last week"*, *"releases this month"*, *"update
 
 Example: *"Find DSH news from the last 3 days"* → agent calls `advanced_search` with `timeRange: "3d"`.
 
+#### Multi-Engine Concurrent Search (`multi_search`)
+
+When you need **cross-source verification** to prevent biases or fail-safes from a single engine, the agent can call `multi_search`:
+
+- **Concurrent Execution**: Defaults to querying the top 3 recommended engines for the query type (or an explicit list passed via `engines`). Each engine independently resolves API keys and handles errors (missing keys are skipped without failing the batch).
+- **Deduplication & Merge**: Normalizes URLs and ranks items by the number of engines that found them (`seenIn` counts), with cross-hit results placed at the top.
+- **Untrusted Boundary**: Enforces the `<untrusted-web-content>` safety wrapper and trims snippets.
+- ⚠️ Note: Running multiple engines in parallel consumes more search quota; use on demand when high source diversity is needed.
+
 #### Fetch Webpage Content (`web_fetch`)
 
 After searching, the agent can **read full webpage content** (e.g., *"Open the first link and summarize it"*). The `web_fetch` tool is enabled by default (official `dsh-web-fetch-http` provider):