Update app.py

app.py

@@ -487,6 +487,7 @@ def health():
         "status": "ok",
         "vectorstore_ready": vectorstore is not None,
         "tools": ["search_primo", "search_pubmed", "search_scholar", "search_consensus", "get_library_info"],
+        "endpoints": ["/rag", "/search", "/agent", "/config", "/year"],
         "models": {
             "gpt": bool(os.environ.get("OPENAI_API_KEY")),
             "claude": bool(os.environ.get("ANTHROPIC_API_KEY")),
@@ -585,6 +586,129 @@ async def rag_query(req: RAGRequest):
         return {"answer": "Error processing your question.", "sources": [], "error": str(e)}
 
 
+# ---- Agent endpoint (Batch 3: tool-calling agent) ----
+@app.post("/agent")
+async def agent_query(req: AgentRequest):
+    """
+    Multi-tool agent. Given a question it:
+    1. Classifies intent (library_info | search_academic | search_medical | general)
+    2. Calls the right combination of tools in parallel
+    3. Synthesises results into a single answer with sources
+    """
+    start = time.time()
+    question = req.question
+    history = [{"role": m.role, "content": m.content} for m in req.history] if req.history else []
+
+    # ---- Step 1: classify intent ----
+    try:
+        classifier_prompt = f"""Classify this library chatbot question into ONE category.
+Question: "{question}"
+Categories:
+- library_info: library hours, services, accounts, policies, databases, staff
+- search_academic: find articles, books, papers on a research topic
+- search_medical: biomedical, clinical, health sciences research
+- general: factual/general knowledge question
+
+Return ONLY valid JSON: {{"intent":"<category>","search_query":"<2-5 keyword search query if searching, else empty>"}}"""
+
+        use_claude = req.model == "claude" and os.environ.get("ANTHROPIC_API_KEY")
+        if use_claude:
+            from langchain_anthropic import ChatAnthropic
+            clf_llm = ChatAnthropic(model="claude-haiku-4-5-20251001", temperature=0, max_tokens=100)
+        else:
+            clf_llm = ChatOpenAI(model="gpt-4o-mini", temperature=0, max_tokens=100)
+
+        clf_resp = clf_llm.invoke(classifier_prompt)
+        clf_text = clf_resp.content.strip()
+        # Extract JSON safely
+        s, e = clf_text.find("{"), clf_text.rfind("}")
+        clf = json.loads(clf_text[s:e+1]) if s != -1 else {}
+        intent = clf.get("intent", "general")
+        search_query = clf.get("search_query", question)
+    except Exception:
+        intent = "general"
+        search_query = question
+
+    # ---- Step 2: run tools based on intent ----
+    tool_results = {}
+    tools_used = []
+
+    if intent == "library_info":
+        # RAG from KU knowledge base
+        rag = await tool_library_info(question, history[-5:] if history else None, model=req.model)
+        tool_results["rag"] = rag
+        tools_used.append("get_library_info")
+
+    elif intent in ("search_academic", "search_medical"):
+        # Run PRIMO + PubMed (if medical) in parallel
+        import asyncio
+        tasks = [tool_search_primo(search_query, limit=5)]
+        if intent == "search_medical":
+            tasks.append(tool_search_pubmed(search_query, limit=3))
+        else:
+            tasks.append(tool_search_scholar(search_query, limit=3))
+
+        raw_results = await asyncio.gather(*tasks, return_exceptions=True)
+        combined = []
+        for r in raw_results:
+            if isinstance(r, dict) and r.get("results"):
+                combined.extend(r["results"])
+                tools_used.append(r.get("source", "unknown"))
+
+        tool_results["search"] = {"results": combined[:8], "total": len(combined)}
+        tools_used = list(set(tools_used))
+
+        # Also get RAG context for library-specific guidance
+        rag = await tool_library_info(question, history[-3:] if history else None, model=req.model)
+        tool_results["rag"] = rag
+        tools_used.append("get_library_info")
+
+    else:
+        # General question — use RAG + LLM
+        rag = await tool_library_info(question, history[-5:] if history else None, model=req.model)
+        tool_results["rag"] = rag
+        tools_used.append("get_library_info")
+
+    # ---- Step 3: synthesise answer ----
+    context_parts = []
+    if "rag" in tool_results and tool_results["rag"].get("answer"):
+        context_parts.append(f"Library Knowledge Base:\n{tool_results['rag']['answer']}")
+    if "search" in tool_results and tool_results["search"].get("results"):
+        top = tool_results["search"]["results"][:3]
+        res_text = "\n".join(f"- {r.get('title','')} by {r.get('creator','')} ({r.get('date','')})" for r in top)
+        context_parts.append(f"Search Results:\n{res_text}")
+
+    synthesis_prompt = f"""You are the Khalifa University Library AI Assistant (Abu Dhabi, UAE). KU = Khalifa University.
+Be concise (3-5 sentences). Include relevant URLs. If search results are present, mention the top 2-3.
+
+Context:
+{chr(10).join(context_parts) if context_parts else 'No additional context.'}
+
+Question: {question}
+Answer:"""
+
+    try:
+        if use_claude:
+            from langchain_anthropic import ChatAnthropic
+            synth_llm = ChatAnthropic(model="claude-haiku-4-5-20251001", temperature=0.2, max_tokens=600)
+        else:
+            synth_llm = ChatOpenAI(model="gpt-4o-mini", temperature=0.2, max_tokens=600)
+        answer = synth_llm.invoke(synthesis_prompt).content
+    except Exception as ex:
+        answer = tool_results.get("rag", {}).get("answer", f"Error: {str(ex)}")
+
+    elapsed = time.time() - start
+    return {
+        "answer": answer,
+        "intent": intent,
+        "tools_used": tools_used,
+        "search_results": tool_results.get("search", {}).get("results", []),
+        "sources": tool_results.get("rag", {}).get("sources", []),
+        "model_used": req.model,
+        "response_time": round(elapsed, 2),
+    }
+
+
 # ---- Rebuild index (protected) ----
 @app.post("/rebuild")
 async def rebuild_index(request: Request):