fix: correlate function calls with execution results

python/packages/autogen-agentchat/src/autogen_agentchat/agents/_assistant_agent.py

@@ -1551,6 +1551,7 @@ async def _execute_tool_call(
                 FunctionExecutionResult(
                     content=f"Error: {e}",
                     call_id=tool_call.id,
+                    uuid=tool_call.uuid,
                     is_error=True,
                     name=tool_call.name,
                 ),
@@ -1568,6 +1569,7 @@ async def _execute_tool_call(
                     FunctionExecutionResult(
                         content=result_as_str,
                         call_id=tool_call.id,
+                        uuid=tool_call.uuid,
                         is_error=False,
                         name=tool_call.name,
                     ),
@@ -1608,6 +1610,7 @@ async def _execute_tool_call(
                     FunctionExecutionResult(
                         content=tool_result.to_text(),
                         call_id=tool_call.id,
+                        uuid=tool_call.uuid,
                         is_error=tool_result.is_error,
                         name=tool_call.name,
                     ),
@@ -1618,6 +1621,7 @@ async def _execute_tool_call(
             FunctionExecutionResult(
                 content=f"Error: tool '{tool_call.name}' not found in any workbench",
                 call_id=tool_call.id,
+                uuid=tool_call.uuid,
                 is_error=True,
                 name=tool_call.name,
             ),

python/packages/autogen-agentchat/tests/test_assistant_agent.py

@@ -237,15 +237,14 @@ async def test_model_client_stream_with_tool_calls() -> None:
             assert isinstance(message.messages[-1], TextMessage)
             assert isinstance(message.messages[1], ToolCallRequestEvent)
             assert message.messages[-1].content == "Example response 2 to task"
-            assert message.messages[1].content == [
-                FunctionCall(id="1", name="_pass_function", arguments=r'{"input": "task"}'),
-                FunctionCall(id="3", name="_echo_function", arguments=r'{"input": "task"}'),
-            ]
+            tool_calls = message.messages[1].content
+            assert [call.id for call in tool_calls] == ["1", "3"]
+            assert [call.name for call in tool_calls] == ["_pass_function", "_echo_function"]
             assert isinstance(message.messages[2], ToolCallExecutionEvent)
-            assert message.messages[2].content == [
-                FunctionExecutionResult(call_id="1", content="pass", is_error=False, name="_pass_function"),
-                FunctionExecutionResult(call_id="3", content="task", is_error=False, name="_echo_function"),
-            ]
+            tool_results = message.messages[2].content
+            assert [result.call_id for result in tool_results] == ["1", "3"]
+            assert [result.content for result in tool_results] == ["pass", "task"]
+            assert [result.uuid for result in tool_results] == [call.uuid for call in tool_calls]
         elif isinstance(message, ModelClientStreamingChunkEvent):
             chunks.append(message.content)
     assert "".join(chunks) == "Example response 2 to task"

python/packages/autogen-core/src/autogen_core/_types.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
-from dataclasses import dataclass
+import uuid
+from dataclasses import dataclass, field
 
 
 @dataclass
@@ -10,3 +11,5 @@ class FunctionCall:
     arguments: str
     # Function to call
     name: str
+    # Framework-generated identifier for correlating the call with its result.
+    uuid: str = field(default_factory=lambda: str(uuid.uuid4()), compare=False)

python/packages/autogen-core/src/autogen_core/models/_types.py

@@ -65,6 +65,9 @@ class FunctionExecutionResult(BaseModel):
     call_id: str
     """The ID of the function call. Note this ID may be empty for some models."""
 
+    uuid: str | None = None
+    """Framework-generated identifier for correlating this result with its function call."""
+
     is_error: bool | None = None
     """Whether the function call resulted in an error."""
 

python/packages/autogen-core/src/autogen_core/tool_agent/_caller_loop.py

@@ -64,7 +64,11 @@ async def tool_agent_caller_loop(
             elif isinstance(result, ToolException):
                 function_results.append(
                     FunctionExecutionResult(
-                        content=f"Error: {result}", call_id=result.call_id, is_error=True, name=result.name
+                        content=f"Error: {result}",
+                        call_id=result.call_id,
+                        is_error=True,
+                        name=result.name,
+                        uuid=result.uuid,
                     )
                 )
             elif isinstance(result, BaseException):

python/packages/autogen-core/src/autogen_core/tool_agent/_tool_agent.py

@@ -1,6 +1,7 @@
 import json
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from typing import List
+from uuid import uuid4
 
 from .. import FunctionCall, MessageContext, RoutedAgent, message_handler
 from ..models import FunctionExecutionResult
@@ -20,6 +21,7 @@ class ToolException(BaseException):
     call_id: str
     content: str
     name: str
+    uuid: str = field(default_factory=lambda: str(uuid4()))
 
 
 @dataclass
@@ -78,7 +80,10 @@ async def handle_function_call(self, message: FunctionCall, ctx: MessageContext)
         tool = next((tool for tool in self._tools if tool.name == message.name), None)
         if tool is None:
             raise ToolNotFoundException(
-                call_id=message.id, content=f"Error: Tool not found: {message.name}", name=message.name
+                call_id=message.id,
+                content=f"Error: Tool not found: {message.name}",
+                name=message.name,
+                uuid=message.uuid,
             )
         else:
             try:
@@ -89,8 +94,15 @@ async def handle_function_call(self, message: FunctionCall, ctx: MessageContext)
                 result_as_str = tool.return_value_as_string(result)
             except json.JSONDecodeError as e:
                 raise InvalidToolArgumentsException(
-                    call_id=message.id, content=f"Error: Invalid arguments: {message.arguments}", name=message.name
+                    call_id=message.id,
+                    content=f"Error: Invalid arguments: {message.arguments}",
+                    name=message.name,
+                    uuid=message.uuid,
                 ) from e
             except Exception as e:
-                raise ToolExecutionException(call_id=message.id, content=f"Error: {e}", name=message.name) from e
-        return FunctionExecutionResult(content=result_as_str, call_id=message.id, is_error=False, name=message.name)
+                raise ToolExecutionException(
+                    call_id=message.id, content=f"Error: {e}", name=message.name, uuid=message.uuid
+                ) from e
+        return FunctionExecutionResult(
+            content=result_as_str, call_id=message.id, is_error=False, name=message.name, uuid=message.uuid
+        )

python/packages/autogen-core/tests/test_tool_agent.py

@@ -1,6 +1,7 @@
 import asyncio
 import json
 import logging
+import uuid
 from typing import Any, AsyncGenerator, List, Literal, Mapping, Optional, Sequence, Union
 
 import pytest
@@ -62,10 +63,9 @@ async def test_tool_agent(caplog: pytest.LogCaptureFixture) -> None:
     runtime.start()
 
     # Test pass function
-    result = await runtime.send_message(
-        FunctionCall(id="1", arguments=json.dumps({"input": "pass"}), name="pass"), agent
-    )
-    assert result == FunctionExecutionResult(call_id="1", content="pass", is_error=False, name="pass")
+    call = FunctionCall(id="1", arguments=json.dumps({"input": "pass"}), name="pass")
+    result = await runtime.send_message(call, agent)
+    assert result == FunctionExecutionResult(call_id="1", content="pass", is_error=False, name="pass", uuid=call.uuid)
 
     # Check log.
     assert any(("ToolCall" in record.message and str(agent) in record.message) for record in caplog.records)
@@ -94,6 +94,15 @@ async def test_tool_agent(caplog: pytest.LogCaptureFixture) -> None:
     await runtime.stop()
 
 
+def test_function_call_has_unique_uuid() -> None:
+    first = FunctionCall(id="1", arguments="{}", name="test")
+    second = FunctionCall(id="1", arguments="{}", name="test")
+
+    assert first.uuid != second.uuid
+    uuid.UUID(first.uuid)
+    uuid.UUID(second.uuid)
+
+
 @pytest.mark.asyncio
 async def test_caller_loop() -> None:
     class MockChatCompletionClient(ChatCompletionClient):

python/packages/autogen-ext/src/autogen_ext/agents/azure/_azure_ai_agent.py

@@ -753,7 +753,11 @@ async def on_messages_stream(
                         is_error = True
                     tool_outputs.append(
                         FunctionExecutionResult(
-                            content=result, call_id=tool_call.id, is_error=is_error, name=tool_call.name
+                            content=result,
+                            call_id=tool_call.id,
+                            uuid=tool_call.uuid,
+                            is_error=is_error,
+                            name=tool_call.name,
                         )
                     )
 

python/packages/autogen-ext/src/autogen_ext/agents/openai/_openai_assistant_agent.py

@@ -467,7 +467,11 @@ async def on_messages_stream(
                         is_error = True
                     tool_outputs.append(
                         FunctionExecutionResult(
-                            content=result, call_id=tool_call.id, is_error=is_error, name=tool_call.name
+                            content=result,
+                            call_id=tool_call.id,
+                            uuid=tool_call.uuid,
+                            is_error=is_error,
+                            name=tool_call.name,
                         )
                     )