From 4ca26e88a5d27a35e4fdf566b652611674ad39d8 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Sat, 23 May 2026 01:29:30 +0000
Subject: [PATCH] fix(gemini realtime): emit function_call_arguments.delta
 before .done

Gemini delivers the full function-call arguments in a single toolCall
frame. The OpenAI Realtime spec orders the streaming events as
output_item.added -> function_call_arguments.delta(+) ->
function_call_arguments.done -> output_item.done. Emit a single delta
carrying the complete arguments string before the matching .done so
spec-compliant SDK clients that accumulate deltas and gate finalisation
on at least one delta arriving do not stall on Gemini tool calls.
---
 .../llms/gemini/realtime/transformation.py    | 18 +++++++++++++
 .../test_gemini_realtime_transformation.py    | 25 +++++++++++--------
 2 files changed, 32 insertions(+), 11 deletions(-)

diff --git a/litellm/llms/gemini/realtime/transformation.py b/litellm/llms/gemini/realtime/transformation.py
index 2a40bab7657..abeb849b05b 100644
--- a/litellm/llms/gemini/realtime/transformation.py
+++ b/litellm/llms/gemini/realtime/transformation.py
@@ -1410,6 +1410,24 @@ class GeminiRealtimeConfig(BaseRealtimeConfig):
                             },
                         )
                     )
+                    # response.function_call_arguments.delta — Gemini delivers
+                    # the full arguments string in a single toolCall frame
+                    # rather than streaming partial chunks, so emit one delta
+                    # carrying the complete payload before the matching
+                    # ``.done`` event. Spec-compliant OpenAI Realtime SDK
+                    # clients accumulate ``delta.delta`` and rely on at least
+                    # one delta before ``.done``.
+                    returned_message.append(
+                        {
+                            "type": "response.function_call_arguments.delta",
+                            "event_id": f"event_{uuid.uuid4()}",
+                            "response_id": current_response_id,
+                            "item_id": item_id,
+                            "output_index": idx,
+                            "call_id": tool_call["call_id"],
+                            "delta": tool_call["arguments"],
+                        }
+                    )
                     # response.function_call_arguments.done
                     returned_message.append(tool_call)
                     # response.output_item.done
diff --git a/tests/test_litellm/llms/gemini/realtime/test_gemini_realtime_transformation.py b/tests/test_litellm/llms/gemini/realtime/test_gemini_realtime_transformation.py
index 1d0ae30f275..d1e596f1359 100644
--- a/tests/test_litellm/llms/gemini/realtime/test_gemini_realtime_transformation.py
+++ b/tests/test_litellm/llms/gemini/realtime/test_gemini_realtime_transformation.py
@@ -735,8 +735,8 @@ def test_gemini_tool_call_emits_response_created_preamble():
     )
 
     responses = result["response"]
-    # Should have: response.created, output_item.added, function_call_arguments.done, output_item.done, conversation.item.created, response.done
-    assert len(responses) >= 6
+    # Should have: response.created, output_item.added, function_call_arguments.delta, function_call_arguments.done, output_item.done, conversation.item.created, response.done
+    assert len(responses) >= 7
     assert responses[0]["type"] == "response.created"
     assert "response" in responses[0]
     assert responses[0]["response"]["status"] == "in_progress"
@@ -749,17 +749,20 @@ def test_gemini_tool_call_emits_response_created_preamble():
     assert responses[1]["type"] == "response.output_item.added"
     assert responses[1]["item"]["type"] == "function_call"
     assert responses[1]["item"]["status"] == "in_progress"
-    assert responses[2]["type"] == "response.function_call_arguments.done"
-    assert responses[3]["type"] == "response.output_item.done"
-    assert responses[3]["item"]["type"] == "function_call"
-    assert responses[3]["item"]["status"] == "completed"
-    assert responses[4]["type"] == "conversation.item.created"
+    assert responses[2]["type"] == "response.function_call_arguments.delta"
+    assert responses[2]["call_id"] == "call_123"
+    assert responses[2]["delta"] == responses[3]["arguments"]
+    assert responses[3]["type"] == "response.function_call_arguments.done"
+    assert responses[4]["type"] == "response.output_item.done"
     assert responses[4]["item"]["type"] == "function_call"
     assert responses[4]["item"]["status"] == "completed"
-    assert responses[5]["type"] == "response.done"
-    assert responses[5]["response"]["status"] == "completed"
-    assert len(responses[5]["response"]["output"]) == 1
-    assert responses[5]["response"]["output"][0]["type"] == "function_call"
+    assert responses[5]["type"] == "conversation.item.created"
+    assert responses[5]["item"]["type"] == "function_call"
+    assert responses[5]["item"]["status"] == "completed"
+    assert responses[6]["type"] == "response.done"
+    assert responses[6]["response"]["status"] == "completed"
+    assert len(responses[6]["response"]["output"]) == 1
+    assert responses[6]["response"]["output"][0]["type"] == "function_call"
     assert result["current_output_item_id"] is None
     assert result["current_response_id"] is None