fix: handle tool_calls as single object or array

- Parse tool_calls whether it's a single object {...} or array [...]
- Normalize to list for consistent processing
- Add debug logging to trace tool execution flow
- Fix variable name (value_str instead of array_str)

src/api/routes.py

@@ -139,19 +139,25 @@ def parse_tool_calls(text: str) -> tuple:
     cleaned_text = re.sub(r'```(?:json)?\s*\n?(.+?)```', r'\1', cleaned_text, flags=re.DOTALL)
     cleaned_text = cleaned_text.strip()
     
-    # Try to find JSON with tool_calls - look for { tool_calls: [...] } pattern
+    # Try to find JSON with tool_calls - look for { tool_calls: [...] } or { tool_calls: {...} } pattern
     try:
         # Look for tool_calls inside braces (handle both quoted and unquoted keys)
-        pattern = r'\{\s*"?tool_calls"?\s*:\s*(\[.*?\])\s*\}'
+        # Match either an array \[...\] or a single object {...}
+        pattern = r'\{\s*"?tool_calls"?\s*:\s*(\[.*?\]|\{.*?\})\s*\}'
         match = re.search(pattern, cleaned_text, re.DOTALL)
         if match:
-            array_str = match.group(1)
+            value_str = match.group(1)
             # Try to parse as JSON first
             try:
-                tool_calls = json.loads(array_str)
+                parsed = json.loads(value_str)
+                # Normalize to list: if it's a dict (single tool), wrap in list
+                if isinstance(parsed, dict):
+                    tool_calls = [parsed]
+                else:
+                    tool_calls = parsed
             except json.JSONDecodeError:
                 # Fix common JSON issues in model output
-                fixed = array_str
+                fixed = value_str
                 # Step 1: Handle unquoted keys (JavaScript style)
                 fixed = re.sub(r'([{,])\s*([a-zA-Z_][a-zA-Z0-9_]*)\s*:', r'\1"\2":', fixed)
                 
@@ -181,7 +187,12 @@ def parse_tool_calls(text: str) -> tuple:
                 fixed = fixed.replace("'", '"')
                 
                 try:
-                    tool_calls = json.loads(fixed)
+                    parsed = json.loads(fixed)
+                    # Normalize to list
+                    if isinstance(parsed, dict):
+                        tool_calls = [parsed]
+                    else:
+                        tool_calls = parsed
                 except json.JSONDecodeError as e2:
                     # If still fails, try one more approach - manual extraction
                     try:
@@ -190,7 +201,7 @@ def parse_tool_calls(text: str) -> tuple:
                         # Find all function blocks - need to handle nested braces
                         # Look for "function": {...} where ... can contain nested braces
                         func_pattern = r'"function":\s*(\{[^{}]*(?:\{[^{}]*\}[^{}]*)*\})'
-                        func_matches = list(re.finditer(func_pattern, array_str, re.DOTALL))
+                        func_matches = list(re.finditer(func_pattern, value_str, re.DOTALL))
                         
                         for i, func_match in enumerate(func_matches):
                             func_content = func_match.group(1)
@@ -367,6 +378,7 @@ async def chat_completions(request: ChatCompletionRequest):
     # Format messages into prompt (with tools if provided)
     prompt = format_messages_with_tools(request.messages, request.tools)
     has_tools = request.tools is not None and len(request.tools) > 0
+    print(f"DEBUG: has_tools={has_tools}, tools_count={len(request.tools) if request.tools else 0}")
     
     # Generate ID
     completion_id = f"chatcmpl-{uuid.uuid4().hex[:12]}"
@@ -507,6 +519,8 @@ async def chat_completions(request: ChatCompletionRequest):
             
             response_text = result.selected_response.text
             tokens_generated = result.selected_response.tokens_generated
+            print(f"DEBUG: Generated response (tokens={tokens_generated})")
+            print(f"DEBUG: Response preview: {response_text[:200]}...")
             
             # Parse tool calls if tools were provided
             content = response_text
@@ -514,7 +528,9 @@ async def chat_completions(request: ChatCompletionRequest):
             finish_reason = "stop"
             
             if has_tools:
+                print(f"DEBUG: Parsing tool calls from response...")
                 content, tool_calls_parsed = parse_tool_calls(response_text)
+                print(f"DEBUG: parse_tool_calls returned: content_len={len(content)}, parsed={tool_calls_parsed is not None}")
                 if tool_calls_parsed:
                     print(f"  🔧 Model requesting {len(tool_calls_parsed)} tool(s)...")
                     executor = get_tool_executor()
@@ -543,6 +559,10 @@ async def chat_completions(request: ChatCompletionRequest):
                     finish_reason = "stop"
                     tool_calls = []  # Clear tool_calls since we executed them
                     print(f"  ✅ All tools executed, returning results")
+                else:
+                    print(f"DEBUG: No tool calls parsed from response")
+            else:
+                print(f"DEBUG: No tools requested, returning normal response")
             
             # Estimate prompt tokens (rough approximation)
             prompt_tokens = len(prompt) // 4