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

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),
+        }
+    )