Agente de web search construido e implementado(Tavily ejemplo de .env en readme)

backend/app/agents/web_search/__init__.py

@@ -0,0 +1,37 @@
+from __future__ import annotations
+
+from .agent import agent
+from .models import WebSearchResponse, WebSearchState
+
+
+async def search_web(
+    query: str,
+    max_results: int = 5,
+    include_raw_content: bool = False,
+) -> WebSearchResponse:
+    """
+    Execute web search using Tavily MCP server.
+
+    Args:
+        query: Search query string
+        max_results: Maximum number of results to return (1-10)
+        include_raw_content: Whether to include full content in results
+
+    Returns:
+        WebSearchResponse with results and summary
+    """
+    state = WebSearchState(
+        user_query=query,
+        max_results=max_results,
+        include_raw_content=include_raw_content,
+    )
+
+    prompt = (
+        f"Search the web for: {query}\n\n"
+        f"Return the top {max_results} most relevant results. "
+        "Provide a concise summary of the key findings."
+    )
+
+    # Ejecutar agente con Tavily API directa
+    result = await agent.run(prompt, deps=state)
+    return result.output

backend/app/agents/web_search/agent.py

@@ -0,0 +1,72 @@
+from __future__ import annotations
+
+from pydantic_ai import Agent, RunContext
+from pydantic_ai.models.openai import OpenAIChatModel
+from pydantic_ai.providers.azure import AzureProvider
+from tavily import TavilyClient
+
+from app.core.config import settings
+
+from .models import WebSearchResponse, WebSearchState, SearchResult
+
+
+provider = AzureProvider(
+    azure_endpoint=settings.AZURE_OPENAI_ENDPOINT,
+    api_version=settings.AZURE_OPENAI_API_VERSION,
+    api_key=settings.AZURE_OPENAI_API_KEY,
+)
+model = OpenAIChatModel(model_name="gpt-4o", provider=provider)
+
+
+tavily_client = TavilyClient(api_key=settings.TAVILY_API_KEY)
+
+agent = Agent(
+    model=model,
+    name="WebSearchAgent",
+    deps_type=WebSearchState,
+    output_type=WebSearchResponse,
+    system_prompt=(
+        "You are a web search assistant powered by Tavily. "
+        "Use the tavily_search tool to find relevant, up-to-date information. "
+        "Return a structured WebSearchResponse with results and a concise summary. "
+        "Always cite your sources with URLs."
+    ),
+)
+
+
+@agent.tool
+def tavily_search(ctx: RunContext[WebSearchState], query: str) -> list[SearchResult]:
+    """Search the web using Tavily API for up-to-date information."""
+    response = tavily_client.search(
+        query=query,
+        max_results=ctx.deps.max_results,
+        search_depth="basic",
+        include_raw_content=ctx.deps.include_raw_content,
+    )
+
+    results = []
+    for item in response.get("results", []):
+        results.append(
+            SearchResult(
+                title=item.get("title", ""),
+                url=item.get("url", ""),
+                content=item.get("content", ""),
+                score=item.get("score"),
+            )
+        )
+
+    return results
+
+
+@agent.output_validator
+def finalize_response(
+    ctx: RunContext[WebSearchState],
+    response: WebSearchResponse,
+) -> WebSearchResponse:
+    """Post-process and validate the search response"""
+    return response.model_copy(
+        update={
+            "query": ctx.deps.user_query,
+            "total_results": len(response.results),
+        }
+    )

backend/app/agents/web_search/models.py

@@ -0,0 +1,29 @@
+from __future__ import annotations
+
+from pydantic import BaseModel, Field
+
+
+class SearchResult(BaseModel):
+    """Individual search result from web search"""
+
+    title: str
+    url: str
+    content: str
+    score: float | None = None
+
+
+class WebSearchState(BaseModel):
+    """State passed to agent tools via deps"""
+
+    user_query: str
+    max_results: int = Field(default=5, ge=1, le=10)
+    include_raw_content: bool = False
+
+
+class WebSearchResponse(BaseModel):
+    """Structured response from the web search agent"""
+
+    query: str
+    results: list[SearchResult]
+    summary: str
+    total_results: int

backend/app/core/config.py

@@ -37,7 +37,7 @@ class Settings(BaseSettings):
     # Azure OpenAI configuración
     AZURE_OPENAI_ENDPOINT: str
     AZURE_OPENAI_API_KEY: str
-    AZURE_OPENAI_API_VERSION: str = "2024-02-01"
+    AZURE_OPENAI_API_VERSION: str = "2024-08-01-preview"
     AZURE_OPENAI_EMBEDDING_MODEL: str = "text-embedding-3-large"
     AZURE_OPENAI_EMBEDDING_DEPLOYMENT: str = "text-embedding-3-large"
 
@@ -57,6 +57,9 @@ class Settings(BaseSettings):
     # LandingAI configuración
     LANDINGAI_API_KEY: str
     LANDINGAI_ENVIRONMENT: str = "production"  # "production" o "eu"
+    
+    TAVILY_API_KEY: str
+    
 
     # Schemas storage
     SCHEMAS_DIR: str = "./data/schemas"

backend/app/routers/agent.py

@@ -6,7 +6,7 @@ from pydantic_ai.ui.vercel_ai import VercelAIAdapter
 from starlette.requests import Request
 from starlette.responses import Response
 
-from app.agents import form_auditor
+from app.agents import form_auditor, web_search
 from app.core.config import settings
 
 provider = AzureProvider(
@@ -361,6 +361,14 @@ async def build_audit_report():
     return result.model_dump()
 
 
+@agent.tool_plain
+async def search_web_information(query: str, max_results: int = 5):
+    """Search the web for up-to-date information using Tavily. Use this when you need current information, news, research, or facts not in your knowledge base."""
+    result = await web_search.search_web(query=query, max_results=max_results)
+
+    return result.model_dump()
+
+
 @router.post("/chat")
 async def chat(request: Request) -> Response:
     return await VercelAIAdapter.dispatch_request(request, agent=agent)