diff --git a/strix/llm/llm.py b/strix/llm/llm.py
index 4f624956..c27ac042 100644
--- a/strix/llm/llm.py
+++ b/strix/llm/llm.py
@@ -202,6 +202,13 @@ class LLM:
 
         accumulated = normalize_tool_format(accumulated)
         accumulated = fix_incomplete_tool_call(_truncate_to_first_function(accumulated))
+
+        thinking_content = ""
+        for match in re.finditer(r"<thinking[^>]*>(.*?)</thinking>", accumulated, re.DOTALL):
+            thinking_content += match.group(1) + "\n"
+        if thinking_content:
+            accumulated = accumulated.replace(thinking_content, "")
+
         yield LLMResponse(
             content=accumulated,
             tool_invocations=parse_tool_invocations(accumulated),
diff --git a/strix/llm/utils.py b/strix/llm/utils.py
index 9771854f..314eee22 100644
--- a/strix/llm/utils.py
+++ b/strix/llm/utils.py
@@ -20,7 +20,7 @@ def normalize_tool_format(content: str) -> str:
       <function="X">                        → <function=X>
       <parameter="X">                       → <parameter=X>
     """
-    if "<invoke" in content or "<function_calls" in content:
+    if "<invoke" in content or "<function_calls" in content or '<parameter name="' in content:
         content = _FUNCTION_CALLS_TAG.sub("", content)
         content = _INVOKE_OPEN.sub(r"<function=\1>", content)
         content = _PARAM_NAME_ATTR.sub(r"<parameter=\1>", content)