BrowseAI-Dev

概述

BrowseAI Dev 是为 AI 代理提供的研究基础设施，提供具有验证功能的网络搜索服务。它实现了一个全面的验证管道，从网页中提取声明并根据实际源文本进行验证。系统提供置信度评分、引文和反幻觉功能。它提供不同的研究深度模式，并支持流式 API 以实现实时反馈。MCP 服务器通过模型上下文协议公开这些功能，使 AI 代理能够访问可靠的研究信息。

试试问 AI

装完之后，这里有 5 个你可以让 AI 做的事：

• search

  Basic web search functionality

  试试:Search for information about quantum computing试试:Find recent news about artificial intelligence
• answer

  Verified search with citations and confidence scores

  试试:Answer what is quantum computing with sources试试:Find information about climate change with verified claims
• extract

  Extract specific information from web pages

  试试:Extract the main findings from this research paper试试:Get the contact information from this company website
• compare

  Compare multiple sources or options

  试试:Compare different approaches to renewable energy试试:Contrast the features of these two smartphone models
• clarity

  Anti-hallucination verification tool

  试试:Verify these claims about AI technology试试:Check if these facts about space exploration are accurate
• session

  Create a persistent research session

  试试:Start a research session about machine learning试试:Create a session to track information about electric vehicles
• feedback

  Submit feedback on search results

  试试:Mark this result as good试试:Flag this claim as incorrect

可对比工具

安装

npm install browseai-dev

pip install browseaidev

pip install langchain-browseaidev

{
  "mcpServers": {
    "browseai-dev": {
      "command": "npx",
      "args": ["-y", "@browseai-dev/mcp-server"]
    }
  }
}

FAQ

验证系统如何工作？

声明从网页中提取并通过原子声明分解、混合检索、语义重排序和多源共识进行验证。每个声明根据验证强度获得置信度评分。

有哪些深度模式？

三种级别：'fast'（单次通过）、'thorough'（多次验证并通过迭代查询）和'deep'（多步骤代理研究，包含差距分析和针对性后续查询）。

BrowseAI-Dev 对比