ai: Ship internal training data.

jarvis.py

@@ -39,6 +39,8 @@ from pptx import Presentation
 
 os.system("apt-get update -q -y && apt-get install -q -y tesseract-ocr tesseract-ocr-eng tesseract-ocr-ind libleptonica-dev libtesseract-dev")
 
+INTERNAL_TRAINING_DATA = os.getenv("INTERNAL_TRAINING_DATA", "")
+
 LINUX_SERVER_HOSTS = [host for host in json.loads(os.getenv("LINUX_SERVER_HOST", "[]")) if host]
 LINUX_SERVER_HOSTS_MARKED = set()
 LINUX_SERVER_HOSTS_ATTEMPTS = {}
@@ -158,7 +160,6 @@ async def fetch_response_async(host, provider_key, selected_model, messages, mod
         raise
 
 async def chat_with_model_async(history, user_input, selected_model_display, sess):
-    global ACTIVE_CANDIDATE
     if not get_available_items(LINUX_SERVER_PROVIDER_KEYS, LINUX_SERVER_PROVIDER_KEYS_MARKED) or not get_available_items(LINUX_SERVER_HOSTS, LINUX_SERVER_HOSTS_MARKED):
         return RESPONSES["RESPONSE_3"]
     if not hasattr(sess, "session_id"):
@@ -167,7 +168,10 @@ async def chat_with_model_async(history, user_input, selected_model_display, ses
     model_config = MODEL_CONFIG.get(selected_model, DEFAULT_CONFIG)
     messages = [{"role": "user", "content": user} for user, _ in history]
     messages += [{"role": "assistant", "content": assistant} for _, assistant in history if assistant]
+    if INTERNAL_TRAINING_DATA:
+        messages.insert(0, {"role": "system", "content": INTERNAL_TRAINING_DATA})
     messages.append({"role": "user", "content": user_input})
+    global ACTIVE_CANDIDATE
     if ACTIVE_CANDIDATE is not None:
         try:
             return await fetch_response_async(ACTIVE_CANDIDATE[0], ACTIVE_CANDIDATE[1], selected_model, messages, model_config, sess.session_id)