Update veryfinal.py

veryfinal.py

@@ -1,4 +1,3 @@
-"""Enhanced LangGraph + Agno Hybrid Agent System"""
 import os
 import time
 import random
@@ -6,160 +5,49 @@ from dotenv import load_dotenv
 from typing import List, Dict, Any, TypedDict, Annotated
 import operator
 
-# LangGraph imports
-from langgraph.graph import START, StateGraph, MessagesState
-from langgraph.prebuilt import tools_condition, ToolNode
-from langgraph.checkpoint.memory import MemorySaver
-
-# LangChain imports
-from langchain_core.messages import SystemMessage, HumanMessage, AIMessage
 from langchain_core.tools import tool
-from langchain_groq import ChatGroq
-from langchain_google_genai import ChatGoogleGenerativeAI
-from langchain_nvidia_ai_endpoints import ChatNVIDIA, NVIDIAEmbeddings
 from langchain_community.tools.tavily_search import TavilySearchResults
-from langchain_community.document_loaders import WikipediaLoader, ArxivLoader, JSONLoader
-from langchain_community.vectorstores import FAISS
+from langchain_community.document_loaders import WikipediaLoader
+from langchain_community.vectorstores import Chroma
 from langchain.tools.retriever import create_retriever_tool
 from langchain_text_splitters import RecursiveCharacterTextSplitter
+from langchain_core.messages import SystemMessage, HumanMessage, AIMessage
+from langchain_community.embeddings import SentenceTransformerEmbeddings
 
-# Agno imports
-from agno.agent import Agent
-from agno.models.groq import GroqChat
-from agno.models.google import GeminiChat
-from agno.tools.tavily import TavilyTools
-from agno.memory.agent import AgentMemory
-from agno.storage.sqlite import SqliteStorage
-from agno.memory.v2.db.sqlite import SqliteMemoryDb # Correct import for memory DB
-
-load_dotenv()
-
-# Rate limiter with exponential backoff
-class PerformanceRateLimiter:
-    def __init__(self, rpm: int, name: str):
-        self.rpm = rpm
-        self.name = name
-        self.times: List[float] = []
-        self.failures = 0
-
-    def wait_if_needed(self):
-        now = time.time()
-        self.times = [t for t in self.times if now - t < 60]
-        if len(self.times) >= self.rpm:
-            wait = 60 - (now - self.times[0]) + random.uniform(1, 3)
-            time.sleep(wait)
-        if self.failures:
-            backoff = min(2 ** self.failures, 30) + random.uniform(0.5, 1.5)
-            time.sleep(backoff)
-        self.times.append(now)
-
-    def record_success(self):
-        self.failures = 0
-
-    def record_failure(self):
-        self.failures += 1
-
-# Initialize rate limiters
-gemini_limiter = PerformanceRateLimiter(28, "Gemini")
-groq_limiter   = PerformanceRateLimiter(28, "Groq")
-nvidia_limiter = PerformanceRateLimiter(4,  "NVIDIA")
-
-# Create Agno agents with corrected SQLite storage and memory
-def create_agno_agents():
-    # 1. Storage for the agent's overall state (conversations, etc.)
-    storage = SqliteStorage(
-        table_name="agent_sessions",
-        db_file="tmp/agent_sessions.db",
-        auto_upgrade_schema=True
-    )
-    # 2. A separate database for the agent's memory
-    memory_db = SqliteMemoryDb(db_file="tmp/agent_memory.db")
-
-    # 3. The AgentMemory object, which uses the memory_db
-    agent_memory = AgentMemory(
-        db=memory_db, # Pass the SqliteMemoryDb here
-        create_user_memories=True,
-        create_session_summary=True
-    )
-
-    math_agent = Agent(
-        name="MathSpecialist",
-        model=GroqChat(
-            model="llama-3.3-70b-versatile",
-            api_key=os.getenv("GROQ_API_KEY"),
-            temperature=0
-        ),
-        description="Expert mathematical problem solver",
-        instructions=[
-            "Solve math problems with precision",
-            "Show step-by-step calculations",
-            "Finish with: FINAL ANSWER: [result]"
-        ],
-        storage=storage,       # Use SqliteStorage for the agent's persistence
-        memory=agent_memory,   # Use the configured AgentMemory
-        show_tool_calls=False,
-        markdown=False
-    )
-    research_agent = Agent(
-        name="ResearchSpecialist",
-        model=GeminiChat(
-            model="gemini-2.0-flash-lite",
-            api_key=os.getenv("GOOGLE_API_KEY"),
-            temperature=0
-        ),
-        description="Expert research and information gathering specialist",
-        instructions=[
-            "Conduct thorough research using available tools",
-            "Synthesize information from multiple sources",
-            "Finish with: FINAL ANSWER: [answer]"
-        ],
-        tools=[
-            TavilyTools(
-                api_key=os.getenv("TAVILY_API_KEY"),
-                search=True,
-                max_tokens=6000,
-                search_depth="advanced",
-                format="markdown"
-            )
-        ],
-        storage=storage,       # Use the same storage for persistence
-        memory=agent_memory,   # Use the same memory configuration
-        show_tool_calls=False,
-        markdown=False
-    )
-    return {"math": math_agent, "research": research_agent}
+from langgraph.graph import StateGraph, START, END
+from langgraph.checkpoint.memory import MemorySaver
 
-# LangGraph tools
+# ---- Tool Definitions ----
 @tool
 def multiply(a: int, b: int) -> int:
-    """Multiply two numbers."""
+    """Multiply two integers and return the product."""
     return a * b
 
 @tool
 def add(a: int, b: int) -> int:
-    """Add two numbers."""
+    """Add two integers and return the sum."""
     return a + b
 
 @tool
 def subtract(a: int, b: int) -> int:
-    """Subtract two numbers."""
+    """Subtract the second integer from the first and return the difference."""
     return a - b
 
 @tool
 def divide(a: int, b: int) -> float:
-    """Divide two numbers."""
+    """Divide the first integer by the second and return the quotient."""
     if b == 0:
         raise ValueError("Cannot divide by zero.")
     return a / b
 
 @tool
 def modulus(a: int, b: int) -> int:
-    """Return the remainder of a divided by b."""
+    """Return the remainder of the division of the first integer by the second."""
     return a % b
 
 @tool
 def optimized_web_search(query: str) -> str:
-    """Optimized Tavily web search."""
+    """Perform an optimized web search using TavilySearchResults and return concatenated document snippets."""
     try:
         time.sleep(random.uniform(1, 2))
         docs = TavilySearchResults(max_results=2).invoke(query=query)
@@ -172,7 +60,7 @@ def optimized_web_search(query: str) -> str:
 
 @tool
 def optimized_wiki_search(query: str) -> str:
-    """Optimized Wikipedia search."""
+    """Perform an optimized Wikipedia search and return concatenated document snippets."""
     try:
         time.sleep(random.uniform(0.5, 1))
         docs = WikipediaLoader(query=query, load_max_docs=1).load()
@@ -183,24 +71,45 @@ def optimized_wiki_search(query: str) -> str:
     except Exception as e:
         return f"Wikipedia search failed: {e}"
 
-# FAISS setup
-def setup_faiss():
+# ---- LLM Integrations ----
+load_dotenv()
+
+# Groq (Llama 3, DeepSeek, etc. via LangChain integration)
+from langchain_groq import ChatGroq
+
+# NVIDIA NIM (LangChain integration)
+from langchain_nvidia_ai_endpoints import ChatNVIDIA
+
+from google import genai
+
+# DeepSeek (via Ollama or API)
+import requests
+
+# Baidu ERNIE (assume open source API, use requests as placeholder)
+def baidu_ernie_generate(prompt, api_key=None):
+    """Call Baidu ERNIE open source API (pseudo-code, replace with actual endpoint and params)."""
+    # Example endpoint and payload for demonstration purposes only:
+    url = "https://api.baidu.com/ernie/v1/generate"
+    headers = {"Authorization": f"Bearer {api_key}"}
+    data = {"model": "ernie-4.5", "prompt": prompt}
     try:
-        schema = """
-        { page_content: .Question, metadata: { task_id: .task_id, Final_answer: ."Final answer" } }
-        """
-        loader = JSONLoader(file_path="metadata.jsonl", jq_schema=schema, json_lines=True, text_content=False)
-        docs = loader.load()
-        splitter = RecursiveCharacterTextSplitter(chunk_size=256, chunk_overlap=50)
-        chunks = splitter.split_documents(docs)
-        embeds = NVIDIAEmbeddings(
-            model="nvidia/nv-embedqa-e5-v5",
-            api_key=os.getenv("NVIDIA_API_KEY")
-        )
-        return FAISS.from_documents(chunks, embeds)
+        resp = requests.post(url, headers=headers, json=data, timeout=30)
+        return resp.json().get("result", "")
+    except Exception as e:
+        return f"ERNIE API error: {e}"
+
+def deepseek_generate(prompt, api_key=None):
+    """Call DeepSeek open source API (pseudo-code, replace with actual endpoint and params)."""
+    url = "https://api.deepseek.com/v1/chat/completions"
+    headers = {"Authorization": f"Bearer {api_key}"}
+    data = {"model": "deepseek-chat", "messages": [{"role": "user", "content": prompt}]}
+    try:
+        resp = requests.post(url, headers=headers, json=data, timeout=30)
+        return resp.json().get("choices", [{}])[0].get("message", {}).get("content", "")
     except Exception as e:
-        print(f"FAISS setup failed: {e}")
-        return None
+        return f"DeepSeek API error: {e}"
+
+# ---- Graph State and System ----
 
 class EnhancedAgentState(TypedDict):
     messages: Annotated[List[HumanMessage|AIMessage], operator.add]
@@ -210,117 +119,95 @@ class EnhancedAgentState(TypedDict):
     perf: Dict[str,Any]
     agno_resp: str
 
-class HybridLangGraphAgnoSystem:
+class HybridLangGraphMultiLLMSystem:
     def __init__(self):
-        self.agno = create_agno_agents()
-        self.store = setup_faiss()
         self.tools = [
             multiply, add, subtract, divide, modulus,
             optimized_web_search, optimized_wiki_search
         ]
-        if self.store:
-            retr = self.store.as_retriever(search_type="similarity", search_kwargs={"k":2})
-            self.tools.append(create_retriever_tool(
-                retriever=retr,
-                name="Question_Search",
-                description="Retrieve similar questions"
-            ))
         self.graph = self._build_graph()
 
     def _build_graph(self):
-        groq_llm = ChatGroq(model="llama-3.3-70b-versatile", temperature=0)
-        gemini_llm = ChatGoogleGenerativeAI(model="gemini-2.0-flash-lite", temperature=0)
-        nvidia_llm = ChatNVIDIA(model="meta/llama-3.1-70b-instruct", temperature=0)
+        groq_llm = ChatGroq(model="llama3-70b-8192", temperature=0, api_key=os.getenv("GROQ_API_KEY"))
+        nvidia_llm = ChatNVIDIA(model="meta/llama3-70b-instruct", temperature=0, api_key=os.getenv("NVIDIA_API_KEY"))
 
         def router(st: EnhancedAgentState) -> EnhancedAgentState:
             q = st["query"].lower()
-            if any(k in q for k in ["calculate","math"]):          t="lg_math"
-            elif any(k in q for k in ["research","analyze"]):      t="agno_research"
-            elif any(k in q for k in ["what is","who is"]):        t="lg_retrieval"
-            else:                                                  t="agno_general"
+            if "groq" in q:      t = "groq"
+            elif "nvidia" in q:  t = "nvidia"
+            elif "gemini" in q or "google" in q: t = "gemini"
+            elif "deepseek" in q: t = "deepseek"
+            elif "ernie" in q or "baidu" in q: t = "baidu"
+            else: t = "groq"  # default
             return {**st, "agent_type": t}
 
-        def lg_math(st: EnhancedAgentState) -> EnhancedAgentState:
-            groq_limiter.wait_if_needed()
-            t0=time.time()
-            llm=groq_llm.bind_tools([multiply,add,subtract,divide,modulus])
-            sys=SystemMessage(content="Fast calculator. FINAL ANSWER: [result]")
-            res=llm.invoke([sys,HumanMessage(content=st["query"])])
-            return {**st, "final_answer":res.content, "perf":{"time":time.time()-t0,"prov":"LG-Groq"}}
-
-        def agno_research(st: EnhancedAgentState) -> EnhancedAgentState:
-            gemini_limiter.wait_if_needed()
-            t0=time.time()
-            resp=self.agno["research"].run(st["query"],stream=False)
-            return {**st, "final_answer":resp, "perf":{"time":time.time()-t0,"prov":"Agno-Gemini"}}
-
-        def lg_retrieval(st: EnhancedAgentState) -> EnhancedAgentState:
-            groq_limiter.wait_if_needed()
-            t0=time.time()
-            llm=groq_llm.bind_tools(self.tools)
-            sys=SystemMessage(content="Retrieve. FINAL ANSWER: [answer]")
-            res=llm.invoke([sys,HumanMessage(content=st["query"])])
-            return {**st, "final_answer":res.content, "perf":{"time":time.time()-t0,"prov":"LG-Retrieval"}}
-
-        def agno_general(st: EnhancedAgentState) -> EnhancedAgentState:
-            nvidia_limiter.wait_if_needed()
-            t0=time.time()
-            if any(k in st["query"].lower() for k in ["calculate","compute"]):
-                resp=self.agno["math"].run(st["query"],stream=False)
-            else:
-                resp=self.agno["research"].run(st["query"],stream=False)
-            return {**st, "final_answer":resp, "perf":{"time":time.time()-t0,"prov":"Agno-General"}}
+        def groq_node(st: EnhancedAgentState) -> EnhancedAgentState:
+            t0 = time.time()
+            sys = SystemMessage(content="Answer as an expert.")
+            res = groq_llm.invoke([sys, HumanMessage(content=st["query"])])
+            return {**st, "final_answer": res.content, "perf": {"time": time.time() - t0, "prov": "Groq"}}
+
+        def nvidia_node(st: EnhancedAgentState) -> EnhancedAgentState:
+            t0 = time.time()
+            sys = SystemMessage(content="Answer as an expert.")
+            res = nvidia_llm.invoke([sys, HumanMessage(content=st["query"])])
+            return {**st, "final_answer": res.content, "perf": {"time": time.time() - t0, "prov": "NVIDIA"}}
+
+        def gemini_node(st: EnhancedAgentState) -> EnhancedAgentState:
+            t0 = time.time()
+            genai.configure(api_key=os.getenv("GEMINI_API_KEY"))
+            model = genai.GenerativeModel("gemini-1.5-pro-latest")
+            res = model.generate_content(st["query"])
+            return {**st, "final_answer": res.text, "perf": {"time": time.time() - t0, "prov": "Gemini"}}
+
+        def deepseek_node(st: EnhancedAgentState) -> EnhancedAgentState:
+            t0 = time.time()
+            resp = deepseek_generate(st["query"], api_key=os.getenv("DEEPSEEK_API_KEY"))
+            return {**st, "final_answer": resp, "perf": {"time": time.time() - t0, "prov": "DeepSeek"}}
+
+        def baidu_node(st: EnhancedAgentState) -> EnhancedAgentState:
+            t0 = time.time()
+            resp = baidu_ernie_generate(st["query"], api_key=os.getenv("BAIDU_API_KEY"))
+            return {**st, "final_answer": resp, "perf": {"time": time.time() - t0, "prov": "ERNIE"}}
 
         def pick(st: EnhancedAgentState) -> str:
             return st["agent_type"]
 
-        g=StateGraph(EnhancedAgentState)
-        g.add_node("router",router)
-        g.add_node("lg_math",lg_math)
-        g.add_node("agno_research",agno_research)
-        g.add_node("lg_retrieval",lg_retrieval)
-        g.add_node("agno_general",agno_general)
+        g = StateGraph(EnhancedAgentState)
+        g.add_node("router", router)
+        g.add_node("groq", groq_node)
+        g.add_node("nvidia", nvidia_node)
+        g.add_node("gemini", gemini_node)
+        g.add_node("deepseek", deepseek_node)
+        g.add_node("baidu", baidu_node)
         g.set_entry_point("router")
-        g.add_conditional_edges("router",pick,{
-            "lg_math":"lg_math","agno_research":"agno_research",
-            "lg_retrieval":"lg_retrieval","agno_general":"agno_general"
+        g.add_conditional_edges("router", pick, {
+            "groq": "groq",
+            "nvidia": "nvidia",
+            "gemini": "gemini",
+            "deepseek": "deepseek",
+            "baidu": "baidu"
         })
-        for n in ["lg_math","agno_research","lg_retrieval","agno_general"]:
-            g.add_edge(n,"END")
+        for n in ["groq", "nvidia", "gemini", "deepseek", "baidu"]:
+            g.add_edge(n, END)
         return g.compile(checkpointer=MemorySaver())
 
-    def process_query(self, q: str) -> Dict[str,Any]:
-        state={
-            "messages":[HumanMessage(content=q)],
-            "query":q,"agent_type":"","final_answer":"",
-            "perf":{},"agno_resp":""
+    def process_query(self, q: str) -> str:
+        state = {
+            "messages": [HumanMessage(content=q)],
+            "query": q,
+            "agent_type": "",
+            "final_answer": "",
+            "perf": {},
+            "agno_resp": ""
         }
-        cfg={"configurable":{"thread_id":f"hyb_{hash(q)}"}}
-        try:
-            out=self.graph.invoke(state,cfg)
-            return {
-                "answer":out["final_answer"],
-                "performance_metrics":out["perf"],
-                "provider_used":out["perf"].get("prov")
-            }
-        except Exception as e:
-            return {"answer":f"Error: {e}","performance_metrics":{},"provider_used":"Error"}
-
-def build_graph(provider: str = "hybrid"):
-    """
-    Build and return the StateGraph for the requested provider.
-    - "hybrid", "groq", "google", and "nvidia" are all valid and
-      will return the full HybridLangGraphAgnoSystem graph.
-    """
-    if provider in ("hybrid", "groq", "google", "nvidia"):
-        return HybridLangGraphAgnoSystem().graph
-    else:
-        raise ValueError(f"Unsupported provider: '{provider}'. Please use 'hybrid', 'groq', 'google', or 'nvidia'.")
-
-# Test
-if __name__=="__main__":
-    graph=build_graph()
-    msgs=[HumanMessage(content="What are the names of the US presidents who were assassinated?")]
-    res=graph.invoke({"messages":msgs},{"configurable":{"thread_id":"test"}})
-    for m in res["messages"]:
-        m.pretty_print()
+        cfg = {"configurable": {"thread_id": f"hyb_{hash(q)}"}}
+        out = self.graph.invoke(state, cfg)
+        raw_answer = out["final_answer"]
+        parts = raw_answer.split('\n\n', 1)
+        answer_part = parts[1].strip() if len(parts) > 1 else raw_answer.strip()
+        return answer_part
+
+if __name__ == "__main__":
+    query = "What are the names of the US presidents who were assassinated? (groq)"
+    print("LangGraph Hybrid:", HybridLangGraphMultiLLMSystem().process_query(query))