Configured llm with together api provider to handle prompts.

agent.py

@@ -1,6 +1,8 @@
+import os
 from typing import TypedDict, Annotated, Optional, Any
 from langgraph.graph.message import add_messages
-from langchain_huggingface import HuggingFaceEndpoint, ChatHuggingFace
+from langchain_openai import ChatOpenAI
+
 from langchain_core.messages import AnyMessage, HumanMessage
 from langgraph.graph import StateGraph, START, END
 from langgraph.prebuilt import ToolNode
@@ -9,6 +11,8 @@ from langchain_core.tools import Tool
 from search_tools import vector_store, arvix_search, question_search, web_search, wiki_search
 from math_tools import add, subtract, modulus, multiply, divide
 
+hf_token = os.environ.get("HF_TOKEN")
+together_token = os.environ.get("TOGETHER_API_TOKEN")
 
 tools = [
     add,
@@ -30,24 +34,27 @@ def build_system_prompt():
 
 class Bariq:
     def __init__(self):
-        base_llm = HuggingFaceEndpoint(
-            repo_id="Qwen/Qwen2.5-32B-Instruct",
-            task="text-generation",
-            max_new_tokens=1024,
-            temperature=0.7,
+        base_llm = ChatOpenAI(
+            model="ServiceNow-AI/Apriel-1.6-15b-Thinker",
+            openai_api_key=together_token,
+            openai_api_base="https://api.together.xyz/v1",
+            max_tokens=1024,
+            temperature=0.1,
+            top_p=0.9,
+            frequency_penalty=0.0,
+            presence_penalty=0.0,
+            stop=["</s>", "User:", "Assistant:", "Final answer:"],
         )
-
-        # Wrap in ChatHuggingFace so it behaves like ChatOpenAI
-        chat_llm = ChatHuggingFace(llm=base_llm)
-
+        
         # Bind your tools exactly like before
-        self.llm = chat_llm.bind_tools(tools)
-
-        # self.llm = ChatOpenAI(model="gpt-5-nano").bind_tools(tools)
+        # self.llm = base_llm.bind(tools=tools)
+        self.llm = base_llm
         
         
 class AgentState(TypedDict):
     messages: Annotated[list[AnyMessage], add_messages]
+    previous_messages: Optional[Annotated[list[AnyMessage], add_messages]]
+    
     
     
 agent = Bariq()
@@ -61,43 +68,31 @@ def assistant(state: AgentState) -> AgentState:
     try:
 
         prompts = []
-        for m in state["messages"]:
-            # print("Message: " + str(m))
+        for m in state["previous_messages"]:
             prompt_content = getattr(m, "content", None)
             message_type = getattr(m, "type", None)
             print("prompt_content " + prompt_content)
-            # print("getattr(m, ""type"", None) " + getattr(m, "type", None))
             
             
-            # if message_type == "human":
-            #     prompts.append({
-            #         "role": "user", 
-            #         "content": prompt_content
-            #         # "content": [{"type": "text", "text": prompt_content}]
-            #     })
-            # elif message_type == "ai":
-            #     prompts.append({
-            #         "role": "assistant", 
-            #         "content": prompt_content
-            #         # "content": [{"type": "text", "text": prompt_content}]
-            #     })
-            # else:
-            #     prompts.append({
-            #         "role": "system", 
-            #         "content": prompt_content
-            #         # "content": [{"type": "text", "text": prompt_content}]
-            #     })
+            if message_type == "human" or message_type == "ai":
+                prompts.append({
+                    "role": "assistant", 
+                    "content": prompt_content
+                })
+            else:
+                prompts.append({
+                    "role": "system", 
+                    "content": prompt_content
+                })
         
         # print("Assistant Node Prompts: " + str(prompts))
-        print("Assistant Node Prompts: " + str(state["messages"]))
         
-        llm_response = agent.llm.invoke(state["messages"])
-
-        print("LLM Response: ", str(llm_response))
-        print("LLM Response content:", llm)
+        llm_response = agent.llm.invoke(prompts)
         
+        print("HumanMessage(content=llm_response.content) " + str(HumanMessage(content=llm_response.content)))
         return {
-            "messages": state["messages"]
+            "previous_messages": state["previous_messages"],
+            "messages": state["previous_messages"] + [HumanMessage(content=llm_response.content)]
         }
     except Exception as e:
         print("Exception in Assistant Node:", e)
@@ -108,30 +103,27 @@ def assistant(state: AgentState) -> AgentState:
 def retriever(state: AgentState) -> AgentState:
     """Retriever Node"""
     try:
-        # print("Retriever Node Messages: " + str(state["messages"]))
-        # for m in state["messages"]:
-        #     print("getattr(m, ""type""): " + getattr(m, "type", None))
+        
         user_messages = [m for m in state["messages"] if getattr(m, "type", None) == "human"]
         query = getattr(user_messages[-1], "content", None)
-        # print("query: " + query)
-
+        
         # Vector search
         similar_docs = vector_store.similarity_search(query, k=2)
 
+        print("similar documents:", str(similar_docs))
         if similar_docs:
             context = similar_docs[0].page_content
-
+            new_messages = state["messages"] + [HumanMessage(content=context)]
             return {
-                "messages": state["messages"] + [HumanMessage(content=context)]
+                "previous_messages": state["messages"],
+                "messages": new_messages
+            }        
+        else:
+            return {
+                "previous_messages": state["messages"] if state["previous_messages"] is None else state["previous_messages"],
+                "messages": state["messages"]
             }
 
-        # print("Retriever response text: ", response_text)
-        
-        
-        return {
-            "messages": state["messages"]
-        }
-
     except Exception as e:
         print("Exception in retriever node:", e)
         

app.py

@@ -35,6 +35,7 @@ def respond(
             
     response_messages = workflow.invoke(            
         {
+            "previous_messages": None,
             "messages": formatted_messages
         }
     )