数据充实代理快速入门

pip install langgraph langchain-brightdata langchain-anthropic

export BRIGHT_DATA_API_KEY="your-bright-data-api-key"
export ANTHROPIC_API_KEY="your-anthropic-api-key"

"""Simple data enrichment agent using LangGraph + Bright Data."""

import json
from dataclasses import dataclass, field
from typing import Any, Annotated, List, Optional

from langchain_anthropic import ChatAnthropic
from langchain_brightdata import BrightDataSERP, BrightDataUnlocker
from langchain_core.messages import AIMessage, HumanMessage, BaseMessage
from langchain_core.tools import tool
from langgraph.graph import StateGraph
from langgraph.graph.message import add_messages
from langgraph.prebuilt import ToolNode


# Agent state tracks topic, schema, conversation history, and final output
@dataclass
class AgentState:
    """State for the enrichment agent."""
    topic: str                                                              # Research topic
    extraction_schema: dict[str, Any]                                       # JSON schema for output
    messages: Annotated[List[BaseMessage], add_messages] = field(default_factory=list)  # Chat history
    info: Optional[dict[str, Any]] = None                                   # Extracted result

# --- Tools ---

# SERP tool: Searches Google and returns structured results
serp_tool = BrightDataSERP(
    search_engine="google",
    country="us",
    language="en",
    results_count=5,
    parse_results=True,
)

# Unlocker tool: Scrapes any URL, bypassing anti-bot protection
unlocker_tool = BrightDataUnlocker(
    data_format="markdown",
)


@tool
async def search(query: str) -> str:
    """Search the web for information about a topic."""
    results = await serp_tool.ainvoke(query)
    return json.dumps(results, indent=2)


@tool
async def scrape_website(url: str) -> str:
    """Scrape and extract content from a specific URL."""
    content = await unlocker_tool.ainvoke(url)
    return str(content)[:20000]  # Limit content to avoid token overflow


tools = [search, scrape_website]

# --- Agent ---

# System prompt instructs the LLM on its research task
SYSTEM_PROMPT = """You are a research agent. Your task is to gather information about a topic and extract structured data.

You have access to these tools:
- search: Search the web for information
- scrape_website: Get content from a specific URL
- submit_info: Call this when you have gathered all the required information

Research topic: {topic}

Required information schema:
{schema}

Search for relevant information, scrape important pages, then call submit_info with the extracted data."""


def create_agent():
    """Create the enrichment agent graph."""
    llm = ChatAnthropic(model="claude-sonnet-4-20250514")

    async def call_model(state: AgentState) -> dict:
        """Call the LLM to decide next action or submit results."""
        prompt = SYSTEM_PROMPT.format(
            topic=state.topic,
            schema=json.dumps(state.extraction_schema, indent=2)
        )
        messages = [HumanMessage(content=prompt)] + list(state.messages)

        # Dynamic tool for structured output submission
        info_tool = {
            "name": "submit_info",
            "description": "Submit the extracted information when done researching.",
            "parameters": state.extraction_schema,
        }

        model = llm.bind_tools(tools + [info_tool])
        response = await model.ainvoke(messages)

        # Check if agent is submitting final info
        info = None
        if hasattr(response, 'tool_calls') and response.tool_calls:
            for tc in response.tool_calls:
                if tc["name"] == "submit_info":
                    info = tc["args"]
                    break

        return {"messages": [response], "info": info}

    def route(state: AgentState) -> str:
        """Route: end if info submitted, else continue tool loop."""
        if state.info:
            return "__end__"
        if not state.messages:
            return "agent"

        last_msg = state.messages[-1]
        if isinstance(last_msg, AIMessage) and hasattr(last_msg, 'tool_calls') and last_msg.tool_calls:
            for tc in last_msg.tool_calls:
                if tc["name"] == "submit_info":
                    return "__end__"
            return "tools"
        return "agent"

    # Build the graph: agent ↔ tools loop until info is extracted
    graph = StateGraph(AgentState)
    graph.add_node("agent", call_model)
    graph.add_node("tools", ToolNode(tools))
    graph.add_edge("__start__", "agent")
    graph.add_conditional_edges("agent", route)
    graph.add_edge("tools", "agent")

    return graph.compile()


async def enrich(topic: str, schema: dict) -> dict:
    """Run the enrichment agent and return structured data."""
    agent = create_agent()
    result = await agent.ainvoke({
        "topic": topic,
        "extraction_schema": schema,
    })
    return result.get("info", {})


# --- Example Usage ---
if __name__ == "__main__":
    import asyncio

    schema = {
        "type": "object",
        "properties": {
            "company_name": {"type": "string"},
            "industry": {"type": "string"},
            "headquarters": {"type": "string"},
            "founded": {"type": "string"},
            "key_products": {"type": "array", "items": {"type": "string"}},
        },
        "required": ["company_name", "industry"]
    }

    result = asyncio.run(enrich("Stripe payments company", schema))
    print(json.dumps(result, indent=2))

python enrichment_agent.py

{
  "company_name": "Stripe",
  "industry": "Financial Technology / Payments",
  "headquarters": "San Francisco, California",
  "founded": "2010",
  "key_products": [
    "Stripe Payments",
    "Stripe Billing",
    "Stripe Connect",
    "Stripe Atlas"
  ]
}

# Extract competitor information
competitor_schema = {
    "type": "object",
    "properties": {
        "competitors": {
            "type": "array",
            "items": {
                "type": "object",
                "properties": {
                    "name": {"type": "string"},
                    "market_position": {"type": "string"},
                    "key_differentiator": {"type": "string"}
                }
            }
        }
    }
}

result = await enrich("Stripe competitors in payment processing", competitor_schema)

serp_tool = BrightDataSERP(
    search_engine="google",
    country="de",      # Germany
    language="de",     # German
    results_count=10,
)

from langchain_openai import ChatOpenAI

# Replace the LLM initialization
llm = ChatOpenAI(model="gpt-4o")