Update app.py

app.py

@@ -1,20 +1,39 @@
-import gradio as gr
-from openai import OpenAI
+import os
+import gradio as gr  
+from openai import OpenAI  
 
+# Load system prompt from environment variable
+SYSTEM_PROMPT = os.getenv("SYSTEM_PROMPT")
 
-client = OpenAI(
-    base_url="http://soggy-sage-goat-8000.1.cricket.hyperbolic.xyz:30000/v1/",
-    api_key="hyperbolic"
-)
+# Initialize client
+client = OpenAI(  
+    base_url="http://soggy-sage-goat-8000.1.cricket.hyperbolic.xyz:30000/v1/",  
+    api_key="hyperbolic"  
+)  
 
-def predict(message, history):
-    history.append({"role": "user", "content": message})
-    stream = client.chat.completions.create(messages=history, model="meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8", stream=True)
-    chunks = []
-    for chunk in stream:
-        chunks.append(chunk.choices[0].delta.content or "")
-        yield "".join(chunks)
+def predict(message, history):  
+    # If history is empty, insert the system prompt
+    if not any(msg["role"] == "system" for msg in history):
+        history.insert(0, {"role": "system", "content": SYSTEM_PROMPT})
 
-demo = gr.ChatInterface(predict, type="messages", title="ACC Emulect+", theme="TejAndrewsACC/Emulect")
+    history.append({"role": "user", "content": message})  
 
-demo.launch()
+    stream = client.chat.completions.create(  
+        messages=history,  
+        model="meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8",  
+        stream=True  
+    )  
+
+    chunks = []  
+    for chunk in stream:  
+        chunks.append(chunk.choices[0].delta.content or "")  
+        yield "".join(chunks)  
+
+demo = gr.ChatInterface(  
+    predict,  
+    type="messages",  
+    title="ACC Emulect+",  
+    theme="TejAndrewsACC/Emulect"  
+)  
+
+demo.launch()