fix(openai): filter duplicate function_call items when using previous_response_id

When using the Responses API with previous_response_id, the delta
input_chat_ctx includes function_call items from the previous LLM
response. The server already knows about these from the referenced
response, so sending them again causes each tool call to appear
twice in the API logs.

Filter out function_call items from the serialized input when
previous_response_id is set, keeping only function_call_output
items which are genuinely new.

Fixes #5136

Author: felmonon

livekit-plugins/livekit-plugins-openai/livekit/plugins/openai/responses/llm.py

@@ -381,6 +381,16 @@ async def _run_impl(self) -> None:
         self._response_completed = False
         chat_ctx, _ = self._chat_ctx.to_provider_format(format="openai.responses")
 
+        # When using previous_response_id, the server already has the
+        # function_call items from that response.  Sending them again
+        # causes each tool call to appear twice in the API logs.
+        if "previous_response_id" in self._extra_kwargs:
+            chat_ctx = [
+                item
+                for item in chat_ctx
+                if not (isinstance(item, dict) and item.get("type") == "function_call")
+            ]
+
         self._tool_ctx = llm.ToolContext(self.tools)
         tool_schemas = cast(
             list[ToolParam],

tests/test_chat_ctx.py

@@ -658,3 +658,47 @@ def test_instructions_as_modality():
     turn2_ctx = turn1_ctx.copy()
     apply_instructions_modality(turn2_ctx, modality="audio")
     assert str(turn2_ctx.items[0].content[0]) == "audio instructions"
+
+
+def test_responses_chat_ctx_excludes_function_calls_with_previous_response_id():
+    """When previous_response_id is set, function_call items from the previous
+    response are already known to the server. Sending them again causes
+    duplicated tool calls in the API logs (issue #5136)."""
+    chat_ctx = ChatContext()
+    chat_ctx.add_message(role="user", content="What is the weather?")
+    chat_ctx.insert(
+        FunctionCall(
+            call_id="call_abc",
+            name="get_weather",
+            arguments='{"city": "SF"}',
+        )
+    )
+    chat_ctx.insert(
+        FunctionCallOutput(
+            call_id="call_abc",
+            name="get_weather",
+            output='{"temp": 72}',
+            is_error=False,
+        )
+    )
+
+    items, _ = chat_ctx.to_provider_format(format="openai.responses")
+
+    # Unfiltered output should contain a function_call item
+    fc_items = [i for i in items if isinstance(i, dict) and i.get("type") == "function_call"]
+    fco_items = [
+        i for i in items if isinstance(i, dict) and i.get("type") == "function_call_output"
+    ]
+    assert len(fc_items) == 1, "expected one function_call item in unfiltered output"
+    assert len(fco_items) == 1, "expected one function_call_output item"
+
+    # Simulate the filter applied in LLMStream._run_impl when
+    # previous_response_id is present
+    filtered = [i for i in items if not (isinstance(i, dict) and i.get("type") == "function_call")]
+
+    fc_after = [i for i in filtered if isinstance(i, dict) and i.get("type") == "function_call"]
+    fco_after = [
+        i for i in filtered if isinstance(i, dict) and i.get("type") == "function_call_output"
+    ]
+    assert len(fc_after) == 0, "function_call items should be removed"
+    assert len(fco_after) == 1, "function_call_output items should be preserved"