Back to Details

ddg-search

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

DDG Search — Usage Guide

DDG Search — 使用指南

When NOT to search: Skip search for well-established facts your training covers reliably (e.g. "what is a list comprehension"). Search when the answer may have changed, when you need citable sources, or when the user explicitly asks you to look something up.
无需搜索的场景：如果你的训练知识已可靠覆盖既定事实（例如“什么是列表推导式”），则无需搜索。当答案可能已更新、你需要可引用的来源，或用户明确要求你查询信息时，再进行搜索。

Tool Selection

工具选择

ToolBest forLatencyCaching
ddg-search_web-search
URLs, snippets, news, source discoveryFast5 min (query string only)
ddg-search_iask-search
Deep questions, academic, analysis, how-toSlow (≤30s)5 min (query+mode+detail)
ddg-search_monica-search
Conversational synthesis, fallbackSlow (≤60s)None — every call hits the API
工具适用场景响应延迟缓存机制
ddg-search_web-search
URL、网页片段、新闻、来源查找快速5分钟（仅基于查询字符串）
ddg-search_iask-search
深度问题、学术研究、分析、操作指南缓慢（≤30秒）5分钟（基于查询+模式+细节程度）
ddg-search_monica-search
对话式合成结果、备选方案缓慢（≤60秒）无缓存 — 每次调用都会请求API

ddg-search_web-search
— Standard Results

ddg-search_web-search
— 标准结果

Parameters:
  • query
    (required): search query in any language
  • mode
    : 
    "short"
    (default) or 
    "detailed"
    • "short"
      → title + URL + snippet (1 HTTP request, fast)
    • "detailed"
      → additionally fetches each result's full page via Jina Reader (headless browser, handles JS-rendered pages). Content returned as cleaned Markdown. Parallel fetches, 8s per URL / 15s total timeout.
  • numResults
    : 1–20, default 3
⚠️ Cache footgun: The cache key is the query string only. If you run 
web-search(query="X", mode="short")
then 
web-search(query="X", mode="detailed")
within 5 minutes, you'll get the cached short results. Request 
mode="detailed"
on the first call. To bust the cache, slightly rephrase the query. To read a single URL directly, use the 
webfetch
tool instead.
参数：
  • query
    （必填）：任意语言的搜索查询词
  • mode
    ：
    "short"
    （默认）或
    "detailed"
    • "short"
      → 标题 + URL + 网页片段（1次HTTP请求，速度快）
    • "detailed"
      → 额外通过Jina Reader（无头浏览器，支持JS渲染页面）获取每个结果的完整页面内容。返回内容为清理后的Markdown格式。并行获取，每个URL耗时8秒 / 总超时时间15秒。
  • numResults
    ：1–20，默认值3
⚠️ 缓存陷阱：缓存键仅为查询字符串。如果在5分钟内先调用
web-search(query="X", mode="short")
，再调用
web-search(query="X", mode="detailed")
，你将获取缓存的简短结果。 首次调用时请指定
mode="detailed"
。若要绕过缓存，可稍微改写查询词。如果要直接读取单个URL，使用
webfetch
工具替代。

ddg-search_iask-search
— AI-Synthesized Answers

ddg-search_iask-search
— AI合成答案

Parameters:
  • query
    (required): natural language question
  • mode
    (default: 
    "thinking"
    at runtime): controls the AI persona 
    • "question"
      — everyday factual questions
    • "academic"
      — scholarly research, citations, scientific topics
    • "forums"
      — community opinions, personal experiences, troubleshooting
    • "wiki"
      — encyclopedic background, concepts, definitions, history
    • "thinking"
      — complex multi-step reasoning and deep analysis
  • detailLevel
    : 
    "concise"
    | 
    "detailed"
    | 
    "comprehensive"
    (default: standard)
参数：
  • query
    （必填）：自然语言问题
  • mode
    （运行时默认值：
    "thinking"
    ）：控制AI角色 
    • "question"
      — 日常事实类问题
    • "academic"
      — 学术研究、引用、科学主题
    • "forums"
      — 社区观点、个人经验、故障排查
    • "wiki"
      — 百科背景、概念、定义、历史
    • "thinking"
      — 复杂多步骤推理和深度分析
  • detailLevel
    ：
    "concise"
    | 
    "detailed"
    | 
    "comprehensive"
    （默认值：标准）

ddg-search_monica-search
— Monica AI Synthesis

ddg-search_monica-search
— Monica AI合成

Parameters:
  • query
    (required): up to 5,000 characters
No mode control. Use when:
  • iask-search
    is slow or unavailable
  • You want a second synthesis perspective
Rate limiting: Monica enforces a 429 rate limit — space out repeated Monica calls. No caching — every call hits the API.
参数：
  • query
    （必填）：最多5000字符
无模式控制。在以下场景使用：
  • iask-search
    速度过慢或不可用时
  • 你需要另一种合成视角时
速率限制：Monica会触发429请求超限限制 — 请间隔重复调用Monica的时间。 无缓存 — 每次调用都会请求API。

Multilingual Queries

多语言查询

Query in the language that yields the best results for the topic:
  • Japanese, Korean, Chinese topics → query in that language
  • Global/tech topics → English often yields richer results
  • For cross-cultural topics, run parallel queries in multiple languages
请选择能为该主题带来最佳结果的语言进行查询：
  • 日语、韩语、中文主题 → 使用对应语言查询
  • 全球/科技主题 → 英语通常能提供更丰富的结果
  • 跨文化主题 → 可同时使用多种语言并行查询

Common Workflows

常见工作流

Quick fact check

快速事实核查

ddg-search_web-search(query="...", numResults=5)
ddg-search_web-search(query="...", numResults=5)

Deep research

深度研究

ddg-search_iask-search(query="...", mode="thinking", detailLevel="comprehensive")
ddg-search_web-search(query="...", numResults=5)
webfetch(url="https://...")
ddg-search_iask-search(query="...", mode="thinking", detailLevel="comprehensive")
ddg-search_web-search(query="...", numResults=5)
webfetch(url="https://...")

Current events / news

时事/新闻

ddg-search_web-search(query="... site:reuters.com OR site:bbc.com", mode="detailed", numResults=5)
ddg-search_web-search(query="... site:reuters.com OR site:bbc.com", mode="detailed", numResults=5)

Technical how-to or academic topic

技术操作指南或学术主题

ddg-search_iask-search(query="...", mode="academic", detailLevel="detailed")
ddg-search_iask-search(query="...", mode="academic", detailLevel="detailed")

Community opinions / troubleshooting

社区观点/故障排查

ddg-search_iask-search(query="...", mode="forums", detailLevel="detailed")
ddg-search_iask-search(query="...", mode="forums", detailLevel="detailed")

When Things Go Wrong

问题排查

  • Empty results → rephrase the query (different keywords, add/remove qualifiers) and retry
  • Monica 429 rate limit → switch to 
    iask-search
    or 
    web-search
    instead
  • iask-search timeout → fall back to 
    monica-search
    or 
    web-search
  • All three fail → tell the user you couldn't retrieve live results and offer to answer from training knowledge with caveats
  • 无搜索结果 → 重新表述查询词（更换关键词、添加/删除限定词）并重试
  • Monica 429请求超限 → 切换为
    iask-search
    或
    web-search
    替代
  • iask-search超时 → fallback到
    monica-search
    或
    web-search
  • 三种工具均失败 → 告知用户无法获取实时结果，并说明可基于训练知识回答，但存在局限性

Citation Format (Required)

引用格式（必填）

Every piece of information sourced from search must be cited inline, at the exact sentence where it appears — not just in a list at the end.
Correct — inline citation:
DeepSeek R1 achieved 79.8% on AIME 2024, matching OpenAI o1 (Epoch AI). Supabase Pro costs $25/month and includes auth, storage, and realtime (Supabase Pricing).
Wrong — end-only citation:
DeepSeek R1 achieved 79.8% on AIME 2024. ... Sources: Epoch AI, Supabase Pricing
The reader should be able to click the link at the point of the claim. A "Sources" section at the end is fine as a summary but never replaces inline citations.
所有来自搜索的信息必须在引用处直接标注，即出现在对应的句子末尾，而不仅仅是在文末列出来源。
正确示例 — 内联引用：
DeepSeek R1在2024年AIME中取得79.8%的成绩，与OpenAI o1持平（Epoch AI）。 Supabase Pro每月费用25美元，包含身份验证、存储和实时功能（Supabase定价）。
错误示例 — 仅文末引用：
DeepSeek R1在2024年AIME中取得79.8%的成绩。 ... 来源：Epoch AI、Supabase定价
读者应能在提出主张的位置点击链接。文末的“来源”部分可作为总结，但绝不能替代内联引用。

Query Tips

查询技巧

  • Be specific: 
    "Python asyncio error handling 2024"
    beats 
    "Python async"
  • Use quotes for exact phrases: 
    "model collapse" definition
  • Add time signals when recency matters: include 
    "2025"
    or 
    "latest"
  • Site filter for trusted sources: 
    query site:docs.python.org
  • Comparisons: use 
    "X vs Y"
    with 
    web-search
    , or 
    iask-search
    with 
    mode="thinking"
  • 表述要具体：
    "Python asyncio error handling 2024"
    比 
    "Python async"
    效果更好
  • 使用引号精确匹配短语：
    "model collapse" definition
  • 当时效性重要时，添加时间标识：包含
    "2025"
    或
    "latest"
  • 站点筛选获取可信来源：
    query site:docs.python.org
  • 对比类查询：使用
    web-search
    并搭配
    "X vs Y"
    ，或使用
    iask-search
    并设置
    mode="thinking"