Upload folder using huggingface_hub

app.py

@@ -1,3 +1,4 @@
+from pydoc import cli
 from dotenv import load_dotenv
 from openai import OpenAI
 from datetime import datetime
@@ -55,10 +56,12 @@ def push(text: str) -> bool:
 
 
 def record_user_details(email, name="Name not provided", notes="not provided"):
+    print(f"Recording {name} with email {email} and notes {notes}")
     push(f"Recording {name} with email {email} and notes {notes}")
     return {"recorded": "ok"}
 
 def record_unknown_question(question):
+    print(f"Recording {question}")
     push(f"Recording {question}")
     return {"recorded": "ok"}
 
@@ -154,12 +157,19 @@ If the user is engaging in discussion, try to steer them towards getting in touc
         ]
 
         def run(client, model):
-            return client.chat.completions.create(
-                model=model,
-                messages=messages,
-                tools=tools,
-                max_tokens=512  # 🔒 hard cap
-            )
+            if client is open_router_client:
+                return client.chat.completions.create(
+                    model=model,
+                    messages=messages,
+                    tools=tools
+                )
+            else:
+                return client.chat.completions.create(
+                    model=model,
+                    messages=messages,
+                    tools=tools,
+                    max_tokens=512  # 🔒 hard cap
+                )
 
         # Clients
         gemini = OpenAI(
@@ -167,12 +177,26 @@ If the user is engaging in discussion, try to steer them towards getting in touc
             base_url="https://generativelanguage.googleapis.com/v1beta/openai/"
         )
         openai_client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
+        open_router_client = OpenAI(
+            api_key=os.getenv("OPEN_ROUTER_API_KEY"),
+            base_url="https://openrouter.ai/api/v1"
+        )
 
         try:
             response = run(gemini, "gemini-2.0-flash")
         except Exception as e:
             if "quota" in str(e).lower() or "resource_exhausted" in str(e).lower():
-                response = run(openai_client, "gpt-4o-mini")
+                print("Google Limit Exceeded! Falling to open API")
+                try:
+                    response = run(openai_client, "gpt-4o-mini")
+                except Exception as gpt:
+                    if "rate limit" in str(gpt).lower() or "rate_limit_exceeded" in str(gpt).lower():
+                        print("Open API rate limit exceeded! Falling to Open Router API")
+                        # xiaomi/mimo-v2-flash:free
+                        # allenai/olmo-3.1-32b-think:free
+                        response = run(open_router_client, "xiaomi/mimo-v2-flash:free")
+                    else:
+                        raise
             else:
                 raise
 
@@ -186,8 +210,7 @@ If the user is engaging in discussion, try to steer them towards getting in touc
             messages.extend(tool_results)
 
             # Final answer (NO tools this time)
-            response = run(gemini, "gemini-2.0-flash")
-
+            response = run(open_router_client, "xiaomi/mimo-v2-flash:free")
         return response.choices[0].message.content