Update tools_engine.py

tools_engine.py

@@ -1,11 +1,6 @@
 """
-Nexari Tools Engine (UPDATED)
-Author: Piyush
-Improvements:
- - Canonical intent labels returned
- - More robust greeting detection
- - Safer web search formatting
- - Defensive error handling
+tools_engine.py - Improved perform_web_search to return structured results with URLs and snippets,
+and canonical intent detection unchanged.
 """
 
 from duckduckgo_search import DDGS
@@ -13,89 +8,66 @@ from transformers import pipeline
 import re
 
 print(">>> Tools: Loading Intent Classification Model...")
-# zero-shot pipeline; if heavy, Docker predownload already handles it
 intent_classifier = pipeline("zero-shot-classification", model="typeform/distilbert-base-uncased-mnli")
 
 def analyze_intent(user_text):
-    """
-    Returns canonical intent labels:
-      - "internet_search"
-      - "coding_request"
-      - "checking_time"
-      - "general"
-    """
     if not user_text:
         return "general"
-
     text_lower = user_text.lower().strip()
-
-    # === HARDCODED SAFETY (Greeting/Identity) ===
     direct_chat_triggers = [
-        "hi", "hello", "hey", "hlo", "hola", "namaste",
-        "what is your name", "who are you", "who r you", "your name"
+        "hi","hello","hey","hlo","namaste",
+        "what is your name", "who are you", "your name"
     ]
-    # exact match or common short greeting at start
     if text_lower in direct_chat_triggers or any(text_lower.startswith(t + " ") for t in direct_chat_triggers):
-        print(f">>> Brain Override: Detected Greeting/Identity -> Force 'general'")
         return "general"
 
-    # === ZERO-SHOT PASSES ===
-    candidate_labels = [
-        "internet search",
-        "general conversation",
-        "coding request",
-        "checking time"
-    ]
-
+    candidate_labels = ["internet search","general conversation","coding request","checking time"]
     try:
         result = intent_classifier(user_text, candidate_labels)
         top_label = result['labels'][0]
         confidence = result['scores'][0]
-        print(f">>> Brain: Detected '{top_label}' ({confidence:.2f})")
-
-        if confidence > 0.5:
-            # map to canonical labels
-            mapping = {
-                "internet search": "internet_search",
-                "general conversation": "general",
-                "coding request": "coding_request",
-                "checking time": "checking_time"
-            }
+        mapping = {
+            "internet search": "internet_search",
+            "general conversation": "general",
+            "coding request": "coding_request",
+            "checking time": "checking_time"
+        }
+        if confidence > 0.45:
             return mapping.get(top_label, "general")
-    except Exception as e:
-        print(f"Intent Error: {e}")
-
+    except Exception:
+        pass
     return "general"
 
-def perform_web_search(user_text, max_results=3):
+def perform_web_search(user_text, max_results=4):
     """
-    Execute a brief web search and return a compact summary.
+    Return structured results:
+    {
+      "query": "...",
+      "results": [
+         {"title": "...", "snippet": "...", "url": "..."},
+         ...
+      ]
+    }
     """
     try:
-        clean_query = user_text.lower()
-        remove_phrases = ["search for", "google", "find", "tell me about", "latest info on", "news about", "lookup"]
-        for phrase in remove_phrases:
-            clean_query = clean_query.replace(phrase, "")
-
-        clean_query = clean_query.strip()
-        if len(clean_query) < 2:
-            clean_query = user_text
-
-        print(f">>> Action: Searching Web for '{clean_query}'...")
-        results = list(DDGS().text(clean_query, max_results=max_results))
+        query = user_text
+        # sanitize small verbs
+        remove_phrases = ["search for","find","google","look up","lookup","what is","tell me"]
+        q = query.lower()
+        for p in remove_phrases:
+            q = q.replace(p, "")
+        q = q.strip() or query
 
-        if results:
-            parts = []
-            for r in results:
-                title = r.get('title', '').strip()
-                body = re.sub(r'\s+', ' ', r.get('body', '').strip())
-                # keep short
-                if len(body) > 250:
-                    body = body[:250].rsplit(' ', 1)[0] + "..."
-                parts.append(f"- {title}: {body}")
-            summary = "\n".join(parts)
-            return f"### WEB DATA (short) ###\n{summary}\nINSTRUCTION: Use this to inform the answer; prefer concise synthesis."
-        return ""
+        results = list(DDGS().text(q, max_results=max_results))
+        structured = {"query": q, "results": []}
+        for r in results:
+            title = r.get("title","").strip()
+            body = re.sub(r'\s+',' ', r.get("body","").strip())
+            url = r.get("href") or r.get("url") or r.get("link") or ""
+            # short snippet
+            snippet = body[:320]
+            structured["results"].append({"title": title, "snippet": snippet, "url": url})
+        return structured
     except Exception as e:
-        print(f"Search Error: {e}")
-        return ""
+        print(f"Search error: {e}")
+        return {"query": user_text, "results": []}