Update app.py

app.py

@@ -178,7 +178,7 @@ def delete_workspace(workspace_id):
 # -------------------------------------------------------------------
 # Main function that calls the Botpress API endpoint
 # -------------------------------------------------------------------
-def chat_with_assistant(user_input, chat_history, bot_id, workspace_id, temperature=0.9, top_p=0.95):
+def chat_with_assistant(user_input, chat_history, bot_id, workspace_id, temperature=0.9, top_p=0.95, max_tokens=None):
     """
     Sends the user input and chat history to the Botpress API endpoint,
     returns the assistant's response and (possibly updated) bot/workspace IDs.
@@ -227,6 +227,10 @@ def chat_with_assistant(user_input, chat_history, bot_id, workspace_id, temperat
         }
     }
 
+    # Add maxTokens to the payload if provided
+    if max_tokens is not None:
+        payload["input"]["maxTokens"] = max_tokens
+
     botpress_url = "https://api.botpress.cloud/v1/chat/actions"
     max_retries = 3
     timeout = 120  # Increased timeout for long messages
@@ -309,6 +313,10 @@ def chat_with_assistant(user_input, chat_history, bot_id, workspace_id, temperat
             else:
                 print(f"Failed to delete bot {bot_id}")
 
+            # Wait 3 seconds before deleting the workspace
+            print("Waiting 3 seconds before deleting workspace...")
+            time.sleep(3)
+
             print(f"Now deleting workspace {workspace_id}...")
             ws_delete_success = delete_workspace(workspace_id)
             if ws_delete_success:
@@ -366,7 +374,8 @@ def chat_endpoint():
          ...
       ],
       "temperature": 0.9,  // Optional, defaults to 0.9
-      "top_p": 0.95        // Optional, defaults to 0.95
+      "top_p": 0.95,       // Optional, defaults to 0.95
+      "max_tokens": 1000   // Optional, defaults to null (no limit)
     }
     Returns JSON with:
     {
@@ -380,9 +389,10 @@ def chat_endpoint():
     user_input = data.get("user_input", "")
     chat_history = data.get("chat_history", [])
 
-    # Get temperature and top_p from request, or use defaults
+    # Get temperature, top_p, and max_tokens from request, or use defaults
     temperature = data.get("temperature", 0.9)
     top_p = data.get("top_p", 0.95)
+    max_tokens = data.get("max_tokens", None)
 
     # Validate temperature and top_p values
     try:
@@ -403,6 +413,17 @@ def chat_endpoint():
         top_p = 0.95
         print(f"Invalid top_p format. Using default: {top_p}")
 
+    # Validate max_tokens if provided
+    if max_tokens is not None:
+        try:
+            max_tokens = int(max_tokens)
+            if max_tokens <= 0:
+                print("Invalid max_tokens value (must be positive). Not using max_tokens.")
+                max_tokens = None
+        except (ValueError, TypeError):
+            print("Invalid max_tokens format. Not using max_tokens.")
+            max_tokens = None
+
     # If we don't yet have a workspace or bot, create them
     if not GLOBAL_WORKSPACE_ID or not GLOBAL_BOT_ID:
         print("No existing IDs found. Creating new workspace and bot...")
@@ -416,7 +437,7 @@ def chat_endpoint():
 
     # Call our function that interacts with Botpress API
     print(f"Sending chat request with existing bot_id={GLOBAL_BOT_ID}, workspace_id={GLOBAL_WORKSPACE_ID}")
-    print(f"Using temperature={temperature}, top_p={top_p}")
+    print(f"Using temperature={temperature}, top_p={top_p}, max_tokens={max_tokens}")
 
     assistant_response, updated_bot_id, updated_workspace_id = chat_with_assistant(
         user_input,
@@ -424,7 +445,8 @@ def chat_endpoint():
         GLOBAL_BOT_ID,
         GLOBAL_WORKSPACE_ID,
         temperature,
-        top_p
+        top_p,
+        max_tokens
     )
 
     # Update global IDs if they changed