Update app.py

app.py

@@ -1,7 +1,8 @@
-# app.py (for Hugging Face Space & local dev)
 import os
 import gradio as gr
 from gradio.components import ChatMessage
+
+# — these are your LangChain imports —
 import openai
 from pydantic import BaseModel
 from langchain_openai import ChatOpenAI
@@ -10,13 +11,13 @@ from langchain_core.output_parsers import PydanticOutputParser
 from langchain.agents import create_tool_calling_agent, AgentExecutor
 from tools import save_tool
 
-# Load local .env only in development; in Spaces rely on HF Secrets
+# load .env in dev
 if os.getenv("HF_SPACE", None) is None:
     from dotenv import load_dotenv
     load_dotenv()
 
-# Retrieve your OpenAI API key from environment (HF Secret or .env)
-openai_api_key = os.environ["OPENAI_API_KEY"]
+OPENAI_API_KEY = os.environ["OPENAI_API_KEY"]
+openai.api_key = OPENAI_API_KEY
 
 class ResearchResponse(BaseModel):
     topic: str
@@ -27,80 +28,68 @@ class ResearchResponse(BaseModel):
     sources: list[str]
     tools_used: list[str]
 
-llm = ChatOpenAI(openai_api_key=openai_api_key, model="gpt-4.1-nano-2025-04-14")
+# 1) set up LLM + parser
+llm = ChatOpenAI(openai_api_key=OPENAI_API_KEY, model="gpt-4o-mini")
 parser = PydanticOutputParser(pydantic_object=ResearchResponse)
 
-prompt = ChatPromptTemplate.from_messages(
-    [
-        ("system", (
-            "You are an empathetic, culturally and religiously sensitive AI assistant specializing in mental health support for Muslim women. "
-            "Your responses must always:\n"
-            "1. Start with a kind acknowledgment of the user's feelings or situation.\n"
-            "2. Offer a gentle validation.\n"
-            "3. Provide a brief, informative summary.\n"
-            "4. Ask a short, open-ended reflection question.\n"
-            "Optionally, include a Quranic verse/tafsir snippet from Al‑Mizan if relevant.\n"
-            "After thinking, return JSON matching the ResearchResponse schema. Don’t output other text.\n{format_instructions}"
-        )),
-        ("placeholder", "{chat_history}"),
-        ("human", "{query}"),
-        ("placeholder", "{agent_scratchpad}"),
-    ]
-).partial(format_instructions=parser.get_format_instructions())
-
-tools = [save_tool]
+# 2) your custom prompt
+prompt = ChatPromptTemplate.from_messages([
+    ("system", (
+        "You are an empathetic, culturally and religiously sensitive AI assistant "
+        "specializing in mental health support for Muslim women.  Always respond:\n"
+        "1. Acknowledge the user’s feelings.\n"
+        "2. Validate gently.\n"
+        "3. Offer a concise informative summary.\n"
+        "4. End with a short open‐ended question.\n"
+        "Optionally cite one Quranic verse/tafsir from Al-Mizan.\n"
+        "Return **only** valid JSON matching the `ResearchResponse` schema:\n"
+        "{format_instructions}"
+    )),
+    ("human", "{query}"),
+])
+prompt = prompt.partial(format_instructions=parser.get_format_instructions())
+
+# 3) wire up your tools + agent
+tools = []
 agent = create_tool_calling_agent(llm=llm, prompt=prompt, tools=tools)
 agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=False)
 
-# Gradio chat interface
-
+# 4) the Gradio chat function
 def chat_fn(user_message, history):
-    # make sure history is a list of ChatMessage
+    # history is List[ChatMessage], or None
     history = history or []
-    # add the user turn
+    # append the user turn as a ChatMessage
     history.append(ChatMessage(role="user", content=user_message))
 
-    result = agent_executor.invoke({
+    # invoke the agent; we pass along the raw chat history as dicts if you need them
+    agent_output = agent_executor.invoke({
         "query": user_message,
-        "chat_history": [ {"role": m.role, "content": m.content} for m in history ]
+        "chat_history": [
+            {"role": m.role, "content": m.content} for m in history
+        ]
     })
+    # parse its JSON output
+    out = parser.parse(agent_output["output"])
 
-    # 3) parse the JSON string back into your schema
-    out = parser.parse(result["output"])
-
-    assistant_msg = f"{out.empathetic_response}\n{out.informative_response}\n{out.quran}\n{out.question}"
-    # add the assistant turn
-    history.append(ChatMessage(role="assistant", content=assistant_msg))
-
-    # return (what the interface shows as the last assistant message, full history)
-    return assistant_msg, history
-
-css = ".chat-message.user {color: #0b5394;} .chat-message.bot {color: #38761d;}"
-
-'''
-def respond(message, history, system_message, max_tokens, temperature, top_p):
-    # history is now a list of {"role":"user"/"assistant","content":...}
-    messages = [{"role":"system","content":system_message}] + history
-    messages.append({"role":"user","content":message})
+    # build the assistant’s final string
+    assistant_content = "\n".join([
+        out.empathetic_response,
+        out.informative_response,
+        out.quran,
+        out.question,
+    ])
 
-    for u,b in history:
-        msgs += [{"role":"user","content":u}, {"role":"assistant","content":b}]
-    messages.append({"role":"user","content":message})
+    # append as ChatMessage
+    history.append(ChatMessage(role="assistant", content=assistant_content))
 
-    resp = ""
-    for chunk in openai.chat.completions.create(
-        model="gpt-4o-mini",      
-        messages=messages,
-        stream=True
-    ):
-        delta = chunk.choices[0].delta.get("content","")
-        resp += delta
-        yield resp'''
+    # return last assistant message (so the UI updates), and the full history
+    return assistant_content, history
 
+# 5) launch ChatInterface
 demo = gr.ChatInterface(
-    fn = chat_fn,
-    title = "Yaqin",
-    type="messages"
+    fn=chat_fn,
+    title="Yaqin",
+    type="messages",   # ensures we expect ChatMessage objects
 )
 
 if __name__ == "__main__":