init

services/search_service.py

@@ -0,0 +1,104 @@
+from typing import List, Optional
+from models.debate import SearchResult, SearchEvidence
+
+
+class SearchService:
+    """
+    Tavily web search wrapper for debate research.
+    """
+
+    def __init__(self, api_key: str):
+        from tavily import TavilyClient
+        self.client = TavilyClient(api_key=api_key)
+
+    def search(self, query: str, max_results: int = 5) -> List[SearchResult]:
+        """
+        Perform a web search using Tavily and return structured results.
+        """
+        try:
+            response = self.client.search(
+                query=query,
+                max_results=max_results,
+                search_depth="basic"
+            )
+            results = []
+            for item in response.get("results", []):
+                results.append(SearchResult(
+                    title=item.get("title", ""),
+                    url=item.get("url", ""),
+                    snippet=item.get("content", "")[:500],  # Truncate to avoid token bloat
+                    score=item.get("score")
+                ))
+            return results
+        except Exception as e:
+            print(f"Tavily search error: {e}")
+            return []
+
+    @staticmethod
+    def format_results_for_context(evidence: SearchEvidence) -> str:
+        """
+        Format search results into a string suitable for injecting into the LLM context.
+        """
+        if not evidence or not evidence.results:
+            return ""
+        lines = [f"\n[网络搜索结果] 搜索词: \"{evidence.query}\""]
+        for i, r in enumerate(evidence.results, 1):
+            lines.append(f"  {i}. {r.title}")
+            lines.append(f"     {r.snippet}")
+            lines.append(f"     来源: {r.url}")
+        lines.append("[搜索结果结束]\n")
+        return "\n".join(lines)
+
+    @staticmethod
+    def generate_search_query(topic: str, last_opponent_argument: Optional[str] = None) -> str:
+        """
+        Generate a search query from the debate topic and the opponent's last argument.
+        """
+        if last_opponent_argument:
+            # Extract key phrases from the opponent's argument (first ~100 chars)
+            snippet = last_opponent_argument[:100].strip()
+            return f"{topic} {snippet}"
+        return topic
+
+    @staticmethod
+    def get_tool_definition() -> dict:
+        """
+        Return the web_search tool definition for function calling.
+        """
+        return {
+            "type": "function",
+            "function": {
+                "name": "web_search",
+                "description": "Search the web for current information relevant to the debate topic. Use this to find facts, statistics, or recent news that support your argument.",
+                "parameters": {
+                    "type": "object",
+                    "properties": {
+                        "query": {
+                            "type": "string",
+                            "description": "The search query to look up"
+                        }
+                    },
+                    "required": ["query"]
+                }
+            }
+        }
+
+    @staticmethod
+    def get_tool_definition_anthropic() -> dict:
+        """
+        Return the web_search tool definition in Anthropic format.
+        """
+        return {
+            "name": "web_search",
+            "description": "Search the web for current information relevant to the debate topic. Use this to find facts, statistics, or recent news that support your argument.",
+            "input_schema": {
+                "type": "object",
+                "properties": {
+                    "query": {
+                        "type": "string",
+                        "description": "The search query to look up"
+                    }
+                },
+                "required": ["query"]
+            }
+        }