fix: capture tool calls in reasoning models

Author: skoob13

posthog/ai/langchain/callbacks.py

@@ -5,6 +5,7 @@
         "Please install LangChain to use this feature: 'pip install langchain'"
     )
 
+import json
 import logging
 import time
 from dataclasses import dataclass
@@ -29,6 +30,7 @@
     HumanMessage,
     SystemMessage,
     ToolMessage,
+    ToolCall,
 )
 from langchain_core.outputs import ChatGeneration, LLMResult
 from pydantic import BaseModel
@@ -630,12 +632,35 @@ def _extract_raw_esponse(last_response):
         return ""
 
 
-def _convert_message_to_dict(message: BaseMessage) -> Dict[str, Any]:
+def _convert_lc_tool_calls_to_oai(
+    tool_calls: list[ToolCall],
+) -> list[dict[str, Any]]:
+    try:
+        return [
+            {
+                "type": "function",
+                "id": tool_call["id"],
+                "function": {
+                    "name": tool_call["name"],
+                    "arguments": json.dumps(tool_call["args"]),
+                },
+            }
+            for tool_call in tool_calls
+        ]
+    except KeyError:
+        return tool_calls
+
+
+def _convert_message_to_dict(message: BaseMessage) -> dict[str, Any]:
     # assistant message
     if isinstance(message, HumanMessage):
         message_dict = {"role": "user", "content": message.content}
     elif isinstance(message, AIMessage):
-        message_dict = {"role": "assistant", "content": message.content}
+        message_dict = {
+            "role": "assistant",
+            "content": message.content,
+            "tool_calls": _convert_lc_tool_calls_to_oai(message.tool_calls),
+        }
     elif isinstance(message, SystemMessage):
         message_dict = {"role": "system", "content": message.content}
     elif isinstance(message, ToolMessage):
@@ -648,6 +673,9 @@ def _convert_message_to_dict(message: BaseMessage) -> Dict[str, Any]:
     if message.additional_kwargs:
         message_dict.update(message.additional_kwargs)
 
+    if "content" in message_dict and not message_dict["content"]:
+        message_dict["content"] = ""
+
     return message_dict
 
 

posthog/test/ai/langchain/test_callbacks.py

@@ -1727,3 +1727,41 @@ def test_openai_reasoning_tokens(mock_client):
     assert call["properties"]["$ai_reasoning_tokens"] is not None
     assert call["properties"]["$ai_input_tokens"] is not None
     assert call["properties"]["$ai_output_tokens"] is not None
+
+
+def test_convert_message_to_dict_tool_calls():
+    """Test that _convert_message_to_dict properly converts tool calls in AIMessage."""
+    from posthog.ai.langchain.callbacks import _convert_message_to_dict
+    from langchain_core.messages import AIMessage
+    from langchain_core.messages.tool import ToolCall
+
+    # Create an AIMessage with tool calls
+    tool_calls = [
+        ToolCall(
+            id="call_123",
+            name="get_weather",
+            args={"city": "San Francisco", "units": "celsius"}
+        )
+    ]
+    
+    ai_message = AIMessage(
+        content="I'll check the weather for you.",
+        tool_calls=tool_calls
+    )
+
+    # Convert to dict
+    result = _convert_message_to_dict(ai_message)
+
+    # Verify the conversion
+    assert result["role"] == "assistant"
+    assert result["content"] == "I'll check the weather for you."
+    assert result["tool_calls"] == [
+        {
+            "type": "function",
+            "id": "call_123",
+            "function": {
+                "name": "get_weather",
+                "arguments": '{"city": "San Francisco", "units": "celsius"}',
+            },
+        }
+    ]

posthog/version.py

@@ -1,4 +1,4 @@
-VERSION = "6.2.1"
+VERSION = "6.2.2"
 
 if __name__ == "__main__":
     print(VERSION, end="")  # noqa: T201