langchain-lastsearch

LangChain integration for LastSearch — verified web search with citations and confidence scores for AI agents.

Unlike raw search APIs, LastSearch fact-checks results before returning them. Every answer includes per-claim verification, cross-source consensus, contradiction detection, and evidence-based confidence scores.

Installation

pip install langchain-lastsearch

Quick Start

from langchain_lastsearch import LastSearchAnswerTool

# Verified search with citations and confidence
tool = LastSearchAnswerTool(api_key="ls_xxx")
result = tool.invoke({"query": "What is quantum computing?"})
print(result)

Available Tools

LastSearchAnswerTool — Verified Research (recommended)

The primary tool. Searches the web, extracts claims, verifies them using multi-signal evidence matching, detects contradictions, and returns an answer with confidence scores.

from langchain_lastsearch import LastSearchAnswerTool

tool = LastSearchAnswerTool(api_key="ls_xxx")

# Fast mode (default)
result = tool.invoke({"query": "Is nuclear energy safe?"})

# Thorough mode (retries if confidence < 60%)
result = tool.invoke({"query": "Health effects of intermittent fasting", "depth": "thorough"})

# Deep mode (multi-step agentic research with gap analysis)
result = tool.invoke({"query": "Compare CRISPR vs base editing approaches", "depth": "deep"})

LastSearchSearchTool — Web Search

Basic web search returning ranked results with URLs, titles, and snippets.

from langchain_lastsearch import LastSearchSearchTool

tool = LastSearchSearchTool(api_key="ls_xxx")
result = tool.invoke({"query": "AI safety regulations 2024", "limit": 5})

LastSearchExtractTool — Page Extraction

Extract structured claims and knowledge from a specific URL.

from langchain_lastsearch import LastSearchExtractTool

tool = LastSearchExtractTool(api_key="ls_xxx")
result = tool.invoke({"url": "https://arxiv.org/abs/2303.08774", "query": "What are GPT-4's capabilities?"})

LastSearchCompareTool — Raw vs Verified

Compare a raw LLM answer against an evidence-backed verified answer. Shows where LLMs hallucinate.

from langchain_lastsearch import LastSearchCompareTool

tool = LastSearchCompareTool(api_key="ls_xxx")
result = tool.invoke({"query": "Is remote work more productive?"})

Use with LangChain Agents

from langchain_lastsearch import LastSearchAnswerTool, LastSearchSearchTool
from langchain_openai import ChatOpenAI
from langchain.agents import create_tool_calling_agent, AgentExecutor
from langchain_core.prompts import ChatPromptTemplate

llm = ChatOpenAI(model="gpt-4o")
tools = [
    LastSearchAnswerTool(api_key="ls_xxx"),
    LastSearchSearchTool(api_key="ls_xxx"),
]

prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a research assistant. Use lastsearch_answer for fact-checked answers with citations."),
    ("human", "{input}"),
    ("placeholder", "{agent_scratchpad}"),
])

agent = create_tool_calling_agent(llm, tools, prompt)
executor = AgentExecutor(agent=agent, tools=tools)

result = executor.invoke({"input": "What are the latest findings on mRNA vaccine efficacy?"})
print(result["output"])

Why LastSearch over Tavily/Exa?

Get an API Key

1. Go to lastsearch.ai
2. Sign in with GitHub
3. Your ls_xxx key is on the dashboard

Your ls_xxx key is on the dashboard.

Links

• Website
• Documentation
• GitHub
• Discord
• Python SDK
• MCP Server

License

Apache 2.0