为人工智能代理提供可靠的研究基础设施。基于证据支持的网络搜索，包含引用、信心评分和Clarity抗幻觉功能。MCP服务器，REST API，Python SDK。
概览
—— 实时网络搜索、证据提取、验证和结构化引用。每个主张都有一个网址支持。每个答案都有一个置信度评分。 npm：lastsearch ·PyPI：lastsearch ·LangChain：langchain-lastsearch — 之前为lastsearch和lastsearch。旧名称仍然有效并自动重定向。 每个答案都经过多重验证流程。没有幻觉。每个说法都有真实来源支持。 置信度评分是——不是LLM自我评估的。在LLM提取主张和来源后，提取后的验证引擎会将每个主张与实际源页文本进行核对： 1. — 复合主张自动拆分为可验证的个体事实。“特斯拉拥有960亿美元收入和180万次交付”变成两个原子主张，各自独立验证。2. — 对于每个主张，关键词匹配通过词汇匹配，密集嵌入则从源文本中寻找语义匹配。排名被整合用于捕捉仅靠关键词匹配未被发现的意译证据（例如：
概述（AI 翻译）
-blue.svg）
基于基础智能的AI代理研究基础设施——实时网页搜索、证据提取、验证和结构化引用。每个主张都有一个网址支持。每个答案都有置信度评分。
网站 ·游乐场 ·API文档 ·替代方案 ·Discord
包名： npm：
lastsearch·PyPI：lastsearch·LangChain：langchain-lastsearch— 之前使用 [lastsearch和lastsearch。旧名称仍然有效并自动重定向。
工作原理
每个答案都经过多重验证流程。没有幻觉。每个说法都有真实来源支持。
验证与信心评分
置信度评分是基于证据的——不是LLM自我评估的。在LLM提取主张和来源后，提取后的验证引擎会将每项主张与实际源页文本进行核对：
- 原子权利要求分解 — 复合声明自动拆分为可验证的个体事实。“特斯拉拥有960亿美元收入和180万交付”变成两个原子权利要求，各自独立验证。
本页由 AI 翻译，技术细节请以原文为准。
README
以下为原文 README 节选，完整内容请查看原文。
LastSearch
-blue.svg){kind=icon}
Research infrastructure for AI agents — real-time web search, evidence extraction, verification against the sources it finds, and structured citations. Claims link to the URLs they came from, and answers carry an estimated confidence score.
Agent → LastSearch → Internet → Answers + sources + estimated confidence
LastSearch is a free, non-commercial passion project by an independent developer. It searches the public internet and weighs what it finds; sources can be wrong, out of date or biased, so nothing it returns is guaranteed to be true or complete. Confidence scores are automated estimates, not certainty, and the final judgement is yours and your agent’s. Not legal, medical, financial or other professional advice. See the Privacy Policy and Terms.
Website · Playground · API Docs · Alternatives · Discord
Package names: npm:
lastsearch· PyPI:lastsearch· LangChain:langchain-lastsearch— Previouslylastsearchandlastsearch. Old names still work and redirect automatically.
How It Works
search → fetch pages → neural rerank → extract claims → verify → cited answer (streamed)
Every answer goes through a multi-step verification pipeline that checks each claim against the source pages it cites, so unsupported claims are flagged instead of passed off as fact.
Verification & Confidence Scoring
Confidence scores are evidence-based — not LLM self-assessed. After the LLM extracts claims and sources, a post-extraction verification engine checks every claim against the actual source page text:
- Atomic claim decomposition — Compound claims are auto-split into individual verifiable facts. “Tesla had $96B revenue and 1.8M deliveries” becomes two atomic claims, each verified independently.
安装
npx -y lastsearch配置
{
"mcpServers": {
"lastsearch": {
"command": "npx",
"args": ["-y", "lastsearch"],
"env": {
"LASTSEARCH_API_KEY": "ls_xxx"
}
}
}
}