From 1b3c8fec83eae73b6caf05d41d1552e5224056f8 Mon Sep 17 00:00:00 2001
From: ".mobo" <1997638+dotmobo@users.noreply.github.com>
Date: Fri, 16 Jan 2026 16:27:09 +0100
Subject: [PATCH 1/9] put logfile and pidfile in /tmp to avoid permission
 denied error in non root environment (#17267)

---
 docker/supervisord.conf | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/docker/supervisord.conf b/docker/supervisord.conf
index c6855fe652..877335804f 100644
--- a/docker/supervisord.conf
+++ b/docker/supervisord.conf
@@ -1,6 +1,8 @@
 [supervisord]
 nodaemon=true
 loglevel=info
+logfile=/tmp/supervisord.log
+pidfile=/tmp/supervisord.pid
 
 [group:litellm]
 programs=main,health

From 2a0f87bde048e406e6dbf1d5465e77d7e0282bc3 Mon Sep 17 00:00:00 2001
From: Dushyant <157586156+dushyantzz@users.noreply.github.com>
Date: Sat, 17 Jan 2026 01:00:16 +0530
Subject: [PATCH 2/9] Fix audio cost per second override (#19158)

---
 docs/my-website/docs/sdk_custom_pricing.md |  5 +++-
 litellm/llms/openai/cost_calculation.py    | 22 ++++++--------
 tests/test_litellm/test_cost_calculator.py | 34 ++++++++++++++++++++++
 3 files changed, 47 insertions(+), 14 deletions(-)

diff --git a/docs/my-website/docs/sdk_custom_pricing.md b/docs/my-website/docs/sdk_custom_pricing.md
index c857711510..ac956db472 100644
--- a/docs/my-website/docs/sdk_custom_pricing.md
+++ b/docs/my-website/docs/sdk_custom_pricing.md
@@ -2,7 +2,10 @@
 
 Register custom pricing for sagemaker completion model. 
 
-For cost per second pricing, you **just** need to register `input_cost_per_second`. 
+For cost per second pricing, register `input_cost_per_second`. If your provider
+charges for audio output duration (e.g., TTS), also set `output_cost_per_second`.
+Values of `0` are treated as not billable, so `output_cost_per_second: 0` will
+not override `input_cost_per_second`.
 
 ```python
 # !pip install boto3 
diff --git a/litellm/llms/openai/cost_calculation.py b/litellm/llms/openai/cost_calculation.py
index e5349db3af..339ce9caed 100644
--- a/litellm/llms/openai/cost_calculation.py
+++ b/litellm/llms/openai/cost_calculation.py
@@ -105,25 +105,21 @@ def cost_per_second(
     prompt_cost = 0.0
     completion_cost = 0.0
     ## Speech / Audio cost calculation
-    if (
-        "output_cost_per_second" in model_info
-        and model_info["output_cost_per_second"] is not None
-    ):
+    output_cost_per_second = model_info.get("output_cost_per_second")
+    input_cost_per_second = model_info.get("input_cost_per_second")
+
+    if output_cost_per_second is not None and output_cost_per_second > 0:
         verbose_logger.debug(
-            f"For model={model} - output_cost_per_second: {model_info.get('output_cost_per_second')}; duration: {duration}"
+            f"For model={model} - output_cost_per_second: {output_cost_per_second}; duration: {duration}"
         )
         ## COST PER SECOND ##
-        completion_cost = model_info["output_cost_per_second"] * duration
-    elif (
-        "input_cost_per_second" in model_info
-        and model_info["input_cost_per_second"] is not None
-    ):
+        completion_cost = output_cost_per_second * duration
+    if input_cost_per_second is not None and input_cost_per_second > 0:
         verbose_logger.debug(
-            f"For model={model} - input_cost_per_second: {model_info.get('input_cost_per_second')}; duration: {duration}"
+            f"For model={model} - input_cost_per_second: {input_cost_per_second}; duration: {duration}"
         )
         ## COST PER SECOND ##
-        prompt_cost = model_info["input_cost_per_second"] * duration
-        completion_cost = 0.0
+        prompt_cost = input_cost_per_second * duration
 
     return prompt_cost, completion_cost
 
diff --git a/tests/test_litellm/test_cost_calculator.py b/tests/test_litellm/test_cost_calculator.py
index 4d6599fc1b..42bcfca231 100644
--- a/tests/test_litellm/test_cost_calculator.py
+++ b/tests/test_litellm/test_cost_calculator.py
@@ -192,6 +192,40 @@ def test_transcription_cost_falls_back_to_duration():
     assert pytest.approx(cost, rel=1e-6) == expected_cost
 
 
+def test_transcription_cost_prefers_input_when_output_zero(monkeypatch):
+    from litellm import completion_cost
+
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
+
+    model_name = "custom-whisper-input-only"
+    custom_model_info = {
+        "input_cost_per_second": 0.00005,
+        "output_cost_per_second": 0.0,
+        "litellm_provider": "openai",
+        "mode": "audio_transcription",
+        "supported_endpoints": ["/v1/audio/transcriptions"],
+    }
+    monkeypatch.setattr(
+        litellm,
+        "model_cost",
+        {**litellm.model_cost, model_name: custom_model_info},
+    )
+
+    response = TranscriptionResponse(text="demo text")
+    response.duration = 300.0
+
+    cost = completion_cost(
+        completion_response=response,
+        model=model_name,
+        custom_llm_provider="openai",
+        call_type="atranscription",
+    )
+
+    expected_cost = 300.0 * 0.00005
+    assert pytest.approx(cost, rel=1e-6) == expected_cost
+
+
 def test_handle_realtime_stream_cost_calculation():
     from litellm.cost_calculator import RealtimeAPITokenUsageProcessor
 

From ce1a2c3209ef46e6f48565e561739cb2579dcbb1 Mon Sep 17 00:00:00 2001
From: Ostap Bodnar <171420608+obod-mpw@users.noreply.github.com>
Date: Sat, 17 Jan 2026 01:18:46 +0200
Subject: [PATCH 3/9] add openai/dall-e base pricing entries (#19133)

---
 model_prices_and_context_window.json | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)

diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json
index a130aefa5d..ec55c251a7 100644
--- a/model_prices_and_context_window.json
+++ b/model_prices_and_context_window.json
@@ -7826,6 +7826,24 @@
         "supports_tool_choice": true,
         "supports_vision": true
     },
+    "dall-e-2": {
+        "input_cost_per_image": 0.02,
+        "litellm_provider": "openai",
+        "mode": "image_generation",
+        "supported_endpoints": [
+            "/v1/images/generations",
+            "/v1/images/edits",
+            "/v1/images/variations"
+        ]
+    },
+    "dall-e-3": {
+        "input_cost_per_image": 0.04,
+        "litellm_provider": "openai",
+        "mode": "image_generation",
+        "supported_endpoints": [
+            "/v1/images/generations"
+        ]
+    },
     "deepseek-chat": {
         "cache_read_input_token_cost": 2.8e-08,
         "input_cost_per_token": 2.8e-07,

From 5a9f6e90cfa7063559bb99a68ed981b1a84cde0a Mon Sep 17 00:00:00 2001
From: Ryan Malloy <ryan@supported.systems>
Date: Fri, 16 Jan 2026 16:24:28 -0700
Subject: [PATCH 4/9] fix(tools): prevent OOM with nested $defs in tool schemas
 (#19098) (#19112)

---
 .../prompt_templates/factory.py               |   7 +-
 litellm/llms/vertex_ai/common_utils.py        |   7 +-
 ...llm_core_utils_prompt_templates_factory.py | 131 ++++++++++++++++++
 3 files changed, 139 insertions(+), 6 deletions(-)

diff --git a/litellm/litellm_core_utils/prompt_templates/factory.py b/litellm/litellm_core_utils/prompt_templates/factory.py
index 89a708077f..6eb4ee7449 100644
--- a/litellm/litellm_core_utils/prompt_templates/factory.py
+++ b/litellm/litellm_core_utils/prompt_templates/factory.py
@@ -4468,9 +4468,10 @@ def _bedrock_tools_pt(tools: List) -> List[BedrockToolBlock]:
 
         defs = parameters.pop("$defs", {})
         defs_copy = copy.deepcopy(defs)
-        # flatten the defs
-        for _, value in defs_copy.items():
-            unpack_defs(value, defs_copy)
+        # Expand $ref references in parameters using the definitions
+        # Note: We don't pre-flatten defs as that causes exponential memory growth
+        # with circular references (see issue #19098). unpack_defs handles nested
+        # refs recursively and correctly detects/skips circular references.
         unpack_defs(parameters, defs_copy)
         tool_input_schema = BedrockToolInputSchemaBlock(
             json=BedrockToolJsonSchemaBlock(
diff --git a/litellm/llms/vertex_ai/common_utils.py b/litellm/llms/vertex_ai/common_utils.py
index 1864ef734c..77c8574228 100644
--- a/litellm/llms/vertex_ai/common_utils.py
+++ b/litellm/llms/vertex_ai/common_utils.py
@@ -453,9 +453,10 @@ def _build_vertex_schema(parameters: dict, add_property_ordering: bool = False):
     valid_schema_fields = set(get_type_hints(Schema).keys())
 
     defs = parameters.pop("$defs", {})
-    # flatten the defs
-    for name, value in defs.items():
-        unpack_defs(value, defs)
+    # Expand $ref references in parameters using the definitions
+    # Note: We don't pre-flatten defs as that causes exponential memory growth
+    # with circular references (see issue #19098). unpack_defs handles nested
+    # refs recursively and correctly detects/skips circular references.
     unpack_defs(parameters, defs)
 
     # 5. Nullable fields:
diff --git a/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py b/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py
index 42a2b5d097..a22fe13798 100644
--- a/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py
+++ b/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py
@@ -1392,3 +1392,134 @@ def test_anthropic_messages_pt_server_tool_use_passthrough():
         b for b in assistant_msg["content"] if b.get("type") == "text"
     )
     assert text_block["text"] == "I found the time tool. How can I help you?"
+
+
+def test_bedrock_tools_unpack_defs_no_oom_with_nested_refs():
+    """
+    Regression test for issue #19098: unpack_defs() causes OOM with nested tool schemas.
+
+    The old implementation had a "flatten defs" loop that would pre-expand each def
+    using unpack_defs(), but since defs often reference each other, each subsequent
+    call would copy already-expanded content, causing exponential memory growth.
+
+    This test creates a schema with multiple nested $defs that reference each other
+    to verify the fix prevents memory explosion while still correctly resolving refs.
+    """
+    import sys
+    import copy
+
+    from litellm.litellm_core_utils.prompt_templates.factory import _bedrock_tools_pt
+
+    # Schema with multiple nested $defs that reference each other
+    # This pattern would cause OOM with the old "flatten defs" loop
+    complex_nested_schema = {
+        "type": "object",
+        "properties": {
+            "query": {"$ref": "#/$defs/Expression"},
+        },
+        "$defs": {
+            "Expression": {
+                "type": "object",
+                "properties": {
+                    "type": {"type": "string", "enum": ["and", "or", "not", "comparison"]},
+                    "left": {"$ref": "#/$defs/Operand"},
+                    "right": {"$ref": "#/$defs/Operand"},
+                    "operator": {"$ref": "#/$defs/Operator"},
+                },
+            },
+            "Operand": {
+                "type": "object",
+                "anyOf": [
+                    {"$ref": "#/$defs/Literal"},
+                    {"$ref": "#/$defs/FieldRef"},
+                    {"$ref": "#/$defs/Expression"},  # Circular: Operand -> Expression -> Operand
+                ],
+            },
+            "Literal": {
+                "type": "object",
+                "properties": {
+                    "type": {"type": "string", "const": "literal"},
+                    "value": {"$ref": "#/$defs/LiteralValue"},
+                },
+            },
+            "LiteralValue": {
+                "oneOf": [
+                    {"type": "string"},
+                    {"type": "number"},
+                    {"type": "boolean"},
+                    {"type": "null"},
+                ],
+            },
+            "FieldRef": {
+                "type": "object",
+                "properties": {
+                    "type": {"type": "string", "const": "field"},
+                    "name": {"type": "string"},
+                    "table": {"$ref": "#/$defs/TableRef"},
+                },
+            },
+            "TableRef": {
+                "type": "object",
+                "properties": {
+                    "name": {"type": "string"},
+                    "alias": {"type": "string"},
+                },
+            },
+            "Operator": {
+                "type": "string",
+                "enum": ["=", "!=", "<", ">", "<=", ">=", "LIKE", "IN"],
+            },
+        },
+    }
+
+    tools = [
+        {
+            "type": "function",
+            "function": {
+                "name": "execute_query",
+                "description": "Execute a query with complex expressions",
+                "parameters": complex_nested_schema,
+            },
+        }
+    ]
+
+    # Measure initial size
+    def get_size(obj, seen=None):
+        size = sys.getsizeof(obj)
+        if seen is None:
+            seen = set()
+        obj_id = id(obj)
+        if obj_id in seen:
+            return 0
+        seen.add(obj_id)
+        if isinstance(obj, dict):
+            size += sum([get_size(v, seen) for v in obj.values()])
+            size += sum([get_size(k, seen) for k in obj.keys()])
+        elif hasattr(obj, "__iter__") and not isinstance(obj, (str, bytes, bytearray)):
+            size += sum([get_size(i, seen) for i in obj])
+        return size
+
+    initial_size = get_size(tools)
+
+    # Process through _bedrock_tools_pt - this should complete without OOM
+    tools_copy = copy.deepcopy(tools)
+    result = _bedrock_tools_pt(tools=tools_copy)
+
+    final_size = get_size(result)
+
+    # The expansion factor should be reasonable (< 100x), not exponential (35000x as in #19098)
+    expansion_factor = final_size / initial_size
+    assert expansion_factor < 100, (
+        f"Memory expansion factor {expansion_factor:.1f}x is too high. "
+        f"Initial: {initial_size} bytes, Final: {final_size} bytes"
+    )
+
+    # Verify the result is valid Bedrock tools format
+    assert isinstance(result, list)
+    assert len(result) == 1
+    assert "toolSpec" in result[0]
+    assert result[0]["toolSpec"]["name"] == "execute_query"
+
+    # Verify $defs have been removed (Bedrock doesn't support them)
+    tool_schema = result[0]["toolSpec"].get("inputSchema", {}).get("json", {})
+    assert "$defs" not in tool_schema, "$defs should be removed after expansion"

From 866bd4674837d64462c70e63fa13343c9ec6a1fb Mon Sep 17 00:00:00 2001
From: Harshit Jain <48647625+Harshit28j@users.noreply.github.com>
Date: Sat, 17 Jan 2026 04:56:52 +0530
Subject: [PATCH 5/9] chore: resolve ModuleNotFoundError for Microsoft Foundry
 Agents (#18991)

---
 litellm/a2a_protocol/main.py                  |  44 ++--
 .../proxy/agent_endpoints/a2a_endpoints.py    |  88 ++++++--
 poetry.lock                                   | 200 ++++++++++--------
 pyproject.toml                                |  16 +-
 requirements.txt                              |  13 +-
 .../agent_endpoints/test_a2a_endpoints.py     |  53 +++--
 6 files changed, 257 insertions(+), 157 deletions(-)

diff --git a/litellm/a2a_protocol/main.py b/litellm/a2a_protocol/main.py
index 167aad7959..2d36dbeacd 100644
--- a/litellm/a2a_protocol/main.py
+++ b/litellm/a2a_protocol/main.py
@@ -113,7 +113,9 @@ def _get_a2a_model_info(a2a_client: Any, kwargs: Dict[str, Any]) -> str:
         litellm_logging_obj.model = model
         litellm_logging_obj.custom_llm_provider = custom_llm_provider
         litellm_logging_obj.model_call_details["model"] = model
-        litellm_logging_obj.model_call_details["custom_llm_provider"] = custom_llm_provider
+        litellm_logging_obj.model_call_details[
+            "custom_llm_provider"
+        ] = custom_llm_provider
 
     return agent_name
 
@@ -197,7 +199,11 @@ async def asend_message(
         )
 
         # Extract params from request
-        params = request.params.model_dump(mode="json") if hasattr(request.params, "model_dump") else dict(request.params)
+        params = (
+            request.params.model_dump(mode="json")
+            if hasattr(request.params, "model_dump")
+            else dict(request.params)
+        )
 
         response_dict = await A2ACompletionBridgeHandler.handle_non_streaming(
             request_id=str(request.id),
@@ -216,7 +222,9 @@ async def asend_message(
     # Create A2A client if not provided but api_base is available
     if a2a_client is None:
         if api_base is None:
-            raise ValueError("Either a2a_client or api_base is required for standard A2A flow")
+            raise ValueError(
+                "Either a2a_client or api_base is required for standard A2A flow"
+            )
         a2a_client = await create_a2a_client(base_url=api_base)
 
     # Type assertion: a2a_client is guaranteed to be non-None here
@@ -235,7 +243,11 @@ async def asend_message(
 
     # Calculate token usage from request and response
     response_dict = a2a_response.model_dump(mode="json", exclude_none=True)
-    prompt_tokens, completion_tokens, _ = A2ARequestUtils.calculate_usage_from_request_response(
+    (
+        prompt_tokens,
+        completion_tokens,
+        _,
+    ) = A2ARequestUtils.calculate_usage_from_request_response(
         request=request,
         response_dict=response_dict,
     )
@@ -280,7 +292,9 @@ def send_message(
     if loop is not None:
         return asend_message(a2a_client=a2a_client, request=request, **kwargs)
     else:
-        return asyncio.run(asend_message(a2a_client=a2a_client, request=request, **kwargs))
+        return asyncio.run(
+            asend_message(a2a_client=a2a_client, request=request, **kwargs)
+        )
 
 
 async def asend_message_streaming(
@@ -347,7 +361,11 @@ async def asend_message_streaming(
         )
 
         # Extract params from request
-        params = request.params.model_dump(mode="json") if hasattr(request.params, "model_dump") else dict(request.params)
+        params = (
+            request.params.model_dump(mode="json")
+            if hasattr(request.params, "model_dump")
+            else dict(request.params)
+        )
 
         async for chunk in A2ACompletionBridgeHandler.handle_streaming(
             request_id=str(request.id),
@@ -365,7 +383,9 @@ async def asend_message_streaming(
     # Create A2A client if not provided but api_base is available
     if a2a_client is None:
         if api_base is None:
-            raise ValueError("Either a2a_client or api_base is required for standard A2A flow")
+            raise ValueError(
+                "Either a2a_client or api_base is required for standard A2A flow"
+            )
         a2a_client = await create_a2a_client(base_url=api_base)
 
     # Type assertion: a2a_client is guaranteed to be non-None here
@@ -378,7 +398,9 @@ async def asend_message_streaming(
     stream = a2a_client.send_message_streaming(request)
 
     # Build logging object for streaming completion callbacks
-    agent_card = getattr(a2a_client, "_litellm_agent_card", None) or getattr(a2a_client, "agent_card", None)
+    agent_card = getattr(a2a_client, "_litellm_agent_card", None) or getattr(
+        a2a_client, "agent_card", None
+    )
     agent_name = getattr(agent_card, "name", "unknown") if agent_card else "unknown"
     model = f"a2a_agent/{agent_name}"
 
@@ -456,7 +478,7 @@ async def create_a2a_client(
     if not A2A_SDK_AVAILABLE:
         raise ImportError(
             "The 'a2a' package is required for A2A agent invocation. "
-            "Install it with: pip install a2a"
+            "Install it with: pip install a2a-sdk"
         )
 
     verbose_logger.info(f"Creating A2A client for {base_url}")
@@ -512,7 +534,7 @@ async def aget_agent_card(
     if not A2A_SDK_AVAILABLE:
         raise ImportError(
             "The 'a2a' package is required for A2A agent invocation. "
-            "Install it with: pip install a2a"
+            "Install it with: pip install a2a-sdk"
         )
 
     verbose_logger.info(f"Fetching agent card from {base_url}")
@@ -534,5 +556,3 @@ async def aget_agent_card(
         f"Fetched agent card: {agent_card.name if hasattr(agent_card, 'name') else 'unknown'}"
     )
     return agent_card
-
-
diff --git a/litellm/proxy/agent_endpoints/a2a_endpoints.py b/litellm/proxy/agent_endpoints/a2a_endpoints.py
index c2d53b40b7..a21f3291d3 100644
--- a/litellm/proxy/agent_endpoints/a2a_endpoints.py
+++ b/litellm/proxy/agent_endpoints/a2a_endpoints.py
@@ -55,9 +55,30 @@ async def _handle_stream_message(
     proxy_server_request: Optional[dict] = None,
 ) -> StreamingResponse:
     """Handle message/stream method via SDK functions."""
-    from a2a.types import MessageSendParams, SendStreamingMessageRequest
-
     from litellm.a2a_protocol import asend_message_streaming
+    from litellm.a2a_protocol.main import A2A_SDK_AVAILABLE
+
+    # Check is handled in invoke_agent_a2a, but if called directly:
+    if not A2A_SDK_AVAILABLE:
+        # Return a streaming response that yields an error
+        async def _error_stream():
+            yield json.dumps(
+                {
+                    "jsonrpc": "2.0",
+                    "id": request_id,
+                    "error": {
+                        "code": -32603,
+                        "message": "Server error: 'a2a' package not installed",
+                    },
+                }
+            ) + "\n"
+
+        return StreamingResponse(_error_stream(), media_type="application/x-ndjson")
+
+    from a2a.types import (
+        MessageSendParams,
+        SendStreamingMessageRequest,
+    )
 
     async def stream_response():
         try:
@@ -75,16 +96,20 @@ async def _handle_stream_message(
             ):
                 # Chunk may be dict or object depending on bridge vs standard path
                 if hasattr(chunk, "model_dump"):
-                    yield json.dumps(chunk.model_dump(mode="json", exclude_none=True)) + "\n"
+                    yield json.dumps(
+                        chunk.model_dump(mode="json", exclude_none=True)
+                    ) + "\n"
                 else:
                     yield json.dumps(chunk) + "\n"
         except Exception as e:
             verbose_proxy_logger.exception(f"Error streaming A2A response: {e}")
-            yield json.dumps({
-                "jsonrpc": "2.0",
-                "id": request_id,
-                "error": {"code": -32603, "message": f"Streaming error: {str(e)}"},
-            }) + "\n"
+            yield json.dumps(
+                {
+                    "jsonrpc": "2.0",
+                    "id": request_id,
+                    "error": {"code": -32603, "message": f"Streaming error: {str(e)}"},
+                }
+            ) + "\n"
 
     return StreamingResponse(stream_response(), media_type="application/x-ndjson")
 
@@ -169,9 +194,8 @@ async def invoke_agent_a2a(
     - message/send: Send a message and get a response
     - message/stream: Send a message and stream the response
     """
-    from a2a.types import MessageSendParams, SendMessageRequest
-
     from litellm.a2a_protocol import asend_message
+    from litellm.a2a_protocol.main import A2A_SDK_AVAILABLE
     from litellm.proxy.agent_endpoints.auth.agent_permission_handler import (
         AgentRequestHandler,
     )
@@ -189,16 +213,28 @@ async def invoke_agent_a2a(
 
         # Validate JSON-RPC format
         if body.get("jsonrpc") != "2.0":
-            return _jsonrpc_error(body.get("id"), -32600, "Invalid Request: jsonrpc must be '2.0'")
+            return _jsonrpc_error(
+                body.get("id"), -32600, "Invalid Request: jsonrpc must be '2.0'"
+            )
 
         request_id = body.get("id")
         method = body.get("method")
         params = body.get("params", {})
 
+        if not A2A_SDK_AVAILABLE:
+            return _jsonrpc_error(
+                request_id,
+                -32603,
+                "Server error: 'a2a' package not installed. Please install 'a2a-sdk'.",
+                500,
+            )
+
         # Find the agent
         agent = _get_agent(agent_id)
         if agent is None:
-            return _jsonrpc_error(request_id, -32000, f"Agent '{agent_id}' not found", 404)
+            return _jsonrpc_error(
+                request_id, -32000, f"Agent '{agent_id}' not found", 404
+            )
 
         is_allowed = await AgentRequestHandler.is_agent_allowed(
             agent_id=agent.agent_id,
@@ -213,23 +249,29 @@ async def invoke_agent_a2a(
         # Get backend URL and agent name
         agent_url = agent.agent_card_params.get("url")
         agent_name = agent.agent_card_params.get("name", agent_id)
-        
+
         # Get litellm_params (may include custom_llm_provider for completion bridge)
         litellm_params = agent.litellm_params or {}
         custom_llm_provider = litellm_params.get("custom_llm_provider")
-        
+
         # URL is required unless using completion bridge with a provider that derives endpoint from model
         # (e.g., bedrock/agentcore derives endpoint from ARN in model string)
         if not agent_url and not custom_llm_provider:
-            return _jsonrpc_error(request_id, -32000, f"Agent '{agent_id}' has no URL configured", 500)
+            return _jsonrpc_error(
+                request_id, -32000, f"Agent '{agent_id}' has no URL configured", 500
+            )
 
-        verbose_proxy_logger.info(f"Proxying A2A request to agent '{agent_id}' at {agent_url or 'completion-bridge'}")
+        verbose_proxy_logger.info(
+            f"Proxying A2A request to agent '{agent_id}' at {agent_url or 'completion-bridge'}"
+        )
 
         # Set up data dict for litellm processing
-        body.update({
-            "model": f"a2a_agent/{agent_name}",
-            "custom_llm_provider": "a2a_agent",
-        })
+        body.update(
+            {
+                "model": f"a2a_agent/{agent_name}",
+                "custom_llm_provider": "a2a_agent",
+            }
+        )
 
         # Add litellm data (user_api_key, user_id, team_id, etc.)
         data = await add_litellm_data_to_request(
@@ -243,6 +285,8 @@ async def invoke_agent_a2a(
 
         # Route through SDK functions
         if method == "message/send":
+            from a2a.types import MessageSendParams, SendMessageRequest
+
             a2a_request = SendMessageRequest(
                 id=request_id,
                 params=MessageSendParams(**params),
@@ -255,7 +299,9 @@ async def invoke_agent_a2a(
                 metadata=data.get("metadata", {}),
                 proxy_server_request=data.get("proxy_server_request"),
             )
-            return JSONResponse(content=response.model_dump(mode="json", exclude_none=True))
+            return JSONResponse(
+                content=response.model_dump(mode="json", exclude_none=True)
+            )
 
         elif method == "message/stream":
             return await _handle_stream_message(
diff --git a/poetry.lock b/poetry.lock
index 3bafdb157c..fd6b28533e 100644
--- a/poetry.lock
+++ b/poetry.lock
@@ -1,4 +1,36 @@
-# This file is automatically @generated by Poetry 2.2.0 and should not be changed by hand.
+# This file is automatically @generated by Poetry 2.2.1 and should not be changed by hand.
+
+[[package]]
+name = "a2a-sdk"
+version = "0.3.22"
+description = "A2A Python SDK"
+optional = true
+python-versions = ">=3.10"
+groups = ["main"]
+markers = "python_version >= \"3.10\" and extra == \"extra-proxy\""
+files = [
+    {file = "a2a_sdk-0.3.22-py3-none-any.whl", hash = "sha256:b98701135bb90b0ff85d35f31533b6b7a299bf810658c1c65f3814a6c15ea385"},
+    {file = "a2a_sdk-0.3.22.tar.gz", hash = "sha256:77a5694bfc4f26679c11b70c7f1062522206d430b34bc1215cfbb1eba67b7e7d"},
+]
+
+[package.dependencies]
+google-api-core = ">=1.26.0"
+httpx = ">=0.28.1"
+httpx-sse = ">=0.4.0"
+protobuf = ">=5.29.5"
+pydantic = ">=2.11.3"
+
+[package.extras]
+all = ["cryptography (>=43.0.0)", "fastapi (>=0.115.2)", "grpcio (>=1.60)", "grpcio-reflection (>=1.7.0)", "grpcio-tools (>=1.60)", "opentelemetry-api (>=1.33.0)", "opentelemetry-sdk (>=1.33.0)", "pyjwt (>=2.0.0)", "sqlalchemy[aiomysql,asyncio] (>=2.0.0)", "sqlalchemy[aiosqlite,asyncio] (>=2.0.0)", "sqlalchemy[asyncio,postgresql-asyncpg] (>=2.0.0)", "sse-starlette", "starlette"]
+encryption = ["cryptography (>=43.0.0)"]
+grpc = ["grpcio (>=1.60)", "grpcio-reflection (>=1.7.0)", "grpcio-tools (>=1.60)"]
+http-server = ["fastapi (>=0.115.2)", "sse-starlette", "starlette"]
+mysql = ["sqlalchemy[aiomysql,asyncio] (>=2.0.0)"]
+postgresql = ["sqlalchemy[asyncio,postgresql-asyncpg] (>=2.0.0)"]
+signing = ["pyjwt (>=2.0.0)"]
+sql = ["sqlalchemy[aiomysql,asyncio] (>=2.0.0)", "sqlalchemy[aiosqlite,asyncio] (>=2.0.0)", "sqlalchemy[asyncio,postgresql-asyncpg] (>=2.0.0)"]
+sqlite = ["sqlalchemy[aiosqlite,asyncio] (>=2.0.0)"]
+telemetry = ["opentelemetry-api (>=1.33.0)", "opentelemetry-sdk (>=1.33.0)"]
 
 [[package]]
 name = "aiofiles"
@@ -1268,25 +1300,6 @@ dev = ["autoflake", "black", "build", "databricks-connect", "httpx", "ipython",
 notebook = ["ipython (>=8,<10)", "ipywidgets (>=8,<9)"]
 openai = ["httpx", "langchain-openai ; python_version > \"3.7\"", "openai"]
 
-[[package]]
-name = "deprecated"
-version = "1.3.1"
-description = "Python @deprecated decorator to deprecate old python classes, functions or methods."
-optional = false
-python-versions = "!=3.0.*,!=3.1.*,!=3.2.*,!=3.3.*,>=2.7"
-groups = ["main", "dev", "proxy-dev"]
-files = [
-    {file = "deprecated-1.3.1-py2.py3-none-any.whl", hash = "sha256:597bfef186b6f60181535a29fbe44865ce137a5079f295b479886c82729d5f3f"},
-    {file = "deprecated-1.3.1.tar.gz", hash = "sha256:b1b50e0ff0c1fddaa5708a2c6b0a6588bb09b892825ab2b214ac9ea9d92a5223"},
-]
-markers = {main = "python_version >= \"3.10\""}
-
-[package.dependencies]
-wrapt = ">=1.10,<3"
-
-[package.extras]
-dev = ["PyTest", "PyTest-Cov", "bump2version (<1)", "setuptools ; python_version >= \"3.12\"", "tox"]
-
 [[package]]
 name = "diskcache"
 version = "5.6.3"
@@ -2521,7 +2534,7 @@ description = "Consume Server-Sent Event (SSE) messages with HTTPX."
 optional = true
 python-versions = ">=3.9"
 groups = ["main"]
-markers = "python_version >= \"3.10\" and extra == \"proxy\""
+markers = "python_version >= \"3.10\" and (extra == \"proxy\" or extra == \"extra-proxy\")"
 files = [
     {file = "httpx_sse-0.4.3-py3-none-any.whl", hash = "sha256:0ac1c9fe3c0afad2e0ebb25a934a59f4c7823b60792691f779fad2c5568830fc"},
     {file = "httpx_sse-0.4.3.tar.gz", hash = "sha256:9b1ed0127459a66014aec3c56bebd93da3c1bc8bb6618c8082039a44889a755d"},
@@ -4036,143 +4049,153 @@ voice-helpers = ["numpy (>=2.0.2)", "sounddevice (>=0.5.1)"]
 
 [[package]]
 name = "opentelemetry-api"
-version = "1.25.0"
+version = "1.39.1"
 description = "OpenTelemetry Python API"
 optional = false
-python-versions = ">=3.8"
+python-versions = ">=3.9"
 groups = ["main", "dev", "proxy-dev"]
 files = [
-    {file = "opentelemetry_api-1.25.0-py3-none-any.whl", hash = "sha256:757fa1aa020a0f8fa139f8959e53dec2051cc26b832e76fa839a6d76ecefd737"},
-    {file = "opentelemetry_api-1.25.0.tar.gz", hash = "sha256:77c4985f62f2614e42ce77ee4c9da5fa5f0bc1e1821085e9a47533a9323ae869"},
+    {file = "opentelemetry_api-1.39.1-py3-none-any.whl", hash = "sha256:2edd8463432a7f8443edce90972169b195e7d6a05500cd29e6d13898187c9950"},
+    {file = "opentelemetry_api-1.39.1.tar.gz", hash = "sha256:fbde8c80e1b937a2c61f20347e91c0c18a1940cecf012d62e65a7caf08967c9c"},
 ]
 markers = {main = "python_version >= \"3.10\""}
 
 [package.dependencies]
-deprecated = ">=1.2.6"
-importlib-metadata = ">=6.0,<=7.1"
+importlib-metadata = ">=6.0,<8.8.0"
+typing-extensions = ">=4.5.0"
 
 [[package]]
 name = "opentelemetry-exporter-otlp"
-version = "1.25.0"
+version = "1.39.1"
 description = "OpenTelemetry Collector Exporters"
 optional = false
-python-versions = ">=3.8"
+python-versions = ">=3.9"
 groups = ["dev", "proxy-dev"]
 files = [
-    {file = "opentelemetry_exporter_otlp-1.25.0-py3-none-any.whl", hash = "sha256:d67a831757014a3bc3174e4cd629ae1493b7ba8d189e8a007003cacb9f1a6b60"},
-    {file = "opentelemetry_exporter_otlp-1.25.0.tar.gz", hash = "sha256:ce03199c1680a845f82e12c0a6a8f61036048c07ec7a0bd943142aca8fa6ced0"},
+    {file = "opentelemetry_exporter_otlp-1.39.1-py3-none-any.whl", hash = "sha256:68ae69775291f04f000eb4b698ff16ff685fdebe5cb52871bc4e87938a7b00fe"},
+    {file = "opentelemetry_exporter_otlp-1.39.1.tar.gz", hash = "sha256:7cf7470e9fd0060c8a38a23e4f695ac686c06a48ad97f8d4867bc9b420180b9c"},
 ]
 
 [package.dependencies]
-opentelemetry-exporter-otlp-proto-grpc = "1.25.0"
-opentelemetry-exporter-otlp-proto-http = "1.25.0"
+opentelemetry-exporter-otlp-proto-grpc = "1.39.1"
+opentelemetry-exporter-otlp-proto-http = "1.39.1"
 
 [[package]]
 name = "opentelemetry-exporter-otlp-proto-common"
-version = "1.25.0"
+version = "1.39.1"
 description = "OpenTelemetry Protobuf encoding"
 optional = false
-python-versions = ">=3.8"
+python-versions = ">=3.9"
 groups = ["dev", "proxy-dev"]
 files = [
-    {file = "opentelemetry_exporter_otlp_proto_common-1.25.0-py3-none-any.whl", hash = "sha256:15637b7d580c2675f70246563363775b4e6de947871e01d0f4e3881d1848d693"},
-    {file = "opentelemetry_exporter_otlp_proto_common-1.25.0.tar.gz", hash = "sha256:c93f4e30da4eee02bacd1e004eb82ce4da143a2f8e15b987a9f603e0a85407d3"},
+    {file = "opentelemetry_exporter_otlp_proto_common-1.39.1-py3-none-any.whl", hash = "sha256:08f8a5862d64cc3435105686d0216c1365dc5701f86844a8cd56597d0c764fde"},
+    {file = "opentelemetry_exporter_otlp_proto_common-1.39.1.tar.gz", hash = "sha256:763370d4737a59741c89a67b50f9e39271639ee4afc999dadfe768541c027464"},
 ]
 
 [package.dependencies]
-opentelemetry-proto = "1.25.0"
+opentelemetry-proto = "1.39.1"
 
 [[package]]
 name = "opentelemetry-exporter-otlp-proto-grpc"
-version = "1.25.0"
+version = "1.39.1"
 description = "OpenTelemetry Collector Protobuf over gRPC Exporter"
 optional = false
-python-versions = ">=3.8"
+python-versions = ">=3.9"
 groups = ["dev", "proxy-dev"]
 files = [
-    {file = "opentelemetry_exporter_otlp_proto_grpc-1.25.0-py3-none-any.whl", hash = "sha256:3131028f0c0a155a64c430ca600fd658e8e37043cb13209f0109db5c1a3e4eb4"},
-    {file = "opentelemetry_exporter_otlp_proto_grpc-1.25.0.tar.gz", hash = "sha256:c0b1661415acec5af87625587efa1ccab68b873745ca0ee96b69bb1042087eac"},
+    {file = "opentelemetry_exporter_otlp_proto_grpc-1.39.1-py3-none-any.whl", hash = "sha256:fa1c136a05c7e9b4c09f739469cbdb927ea20b34088ab1d959a849b5cc589c18"},
+    {file = "opentelemetry_exporter_otlp_proto_grpc-1.39.1.tar.gz", hash = "sha256:772eb1c9287485d625e4dbe9c879898e5253fea111d9181140f51291b5fec3ad"},
 ]
 
 [package.dependencies]
-deprecated = ">=1.2.6"
-googleapis-common-protos = ">=1.52,<2.0"
-grpcio = ">=1.0.0,<2.0.0"
+googleapis-common-protos = ">=1.57,<2.0"
+grpcio = [
+    {version = ">=1.63.2,<2.0.0", markers = "python_version < \"3.13\""},
+    {version = ">=1.66.2,<2.0.0", markers = "python_version >= \"3.13\""},
+]
 opentelemetry-api = ">=1.15,<2.0"
-opentelemetry-exporter-otlp-proto-common = "1.25.0"
-opentelemetry-proto = "1.25.0"
-opentelemetry-sdk = ">=1.25.0,<1.26.0"
+opentelemetry-exporter-otlp-proto-common = "1.39.1"
+opentelemetry-proto = "1.39.1"
+opentelemetry-sdk = ">=1.39.1,<1.40.0"
+typing-extensions = ">=4.6.0"
+
+[package.extras]
+gcp-auth = ["opentelemetry-exporter-credential-provider-gcp (>=0.59b0)"]
 
 [[package]]
 name = "opentelemetry-exporter-otlp-proto-http"
-version = "1.25.0"
+version = "1.39.1"
 description = "OpenTelemetry Collector Protobuf over HTTP Exporter"
 optional = false
-python-versions = ">=3.8"
+python-versions = ">=3.9"
 groups = ["dev", "proxy-dev"]
 files = [
-    {file = "opentelemetry_exporter_otlp_proto_http-1.25.0-py3-none-any.whl", hash = "sha256:2eca686ee11b27acd28198b3ea5e5863a53d1266b91cda47c839d95d5e0541a6"},
-    {file = "opentelemetry_exporter_otlp_proto_http-1.25.0.tar.gz", hash = "sha256:9f8723859e37c75183ea7afa73a3542f01d0fd274a5b97487ea24cb683d7d684"},
+    {file = "opentelemetry_exporter_otlp_proto_http-1.39.1-py3-none-any.whl", hash = "sha256:d9f5207183dd752a412c4cd564ca8875ececba13be6e9c6c370ffb752fd59985"},
+    {file = "opentelemetry_exporter_otlp_proto_http-1.39.1.tar.gz", hash = "sha256:31bdab9745c709ce90a49a0624c2bd445d31a28ba34275951a6a362d16a0b9cb"},
 ]
 
 [package.dependencies]
-deprecated = ">=1.2.6"
 googleapis-common-protos = ">=1.52,<2.0"
 opentelemetry-api = ">=1.15,<2.0"
-opentelemetry-exporter-otlp-proto-common = "1.25.0"
-opentelemetry-proto = "1.25.0"
-opentelemetry-sdk = ">=1.25.0,<1.26.0"
+opentelemetry-exporter-otlp-proto-common = "1.39.1"
+opentelemetry-proto = "1.39.1"
+opentelemetry-sdk = ">=1.39.1,<1.40.0"
 requests = ">=2.7,<3.0"
+typing-extensions = ">=4.5.0"
+
+[package.extras]
+gcp-auth = ["opentelemetry-exporter-credential-provider-gcp (>=0.59b0)"]
 
 [[package]]
 name = "opentelemetry-proto"
-version = "1.25.0"
+version = "1.39.1"
 description = "OpenTelemetry Python Proto"
 optional = false
-python-versions = ">=3.8"
+python-versions = ">=3.9"
 groups = ["main", "dev", "proxy-dev"]
 files = [
-    {file = "opentelemetry_proto-1.25.0-py3-none-any.whl", hash = "sha256:f07e3341c78d835d9b86665903b199893befa5e98866f63d22b00d0b7ca4972f"},
-    {file = "opentelemetry_proto-1.25.0.tar.gz", hash = "sha256:35b6ef9dc4a9f7853ecc5006738ad40443701e52c26099e197895cbda8b815a3"},
+    {file = "opentelemetry_proto-1.39.1-py3-none-any.whl", hash = "sha256:22cdc78efd3b3765d09e68bfbd010d4fc254c9818afd0b6b423387d9dee46007"},
+    {file = "opentelemetry_proto-1.39.1.tar.gz", hash = "sha256:6c8e05144fc0d3ed4d22c2289c6b126e03bcd0e6a7da0f16cedd2e1c2772e2c8"},
 ]
 markers = {main = "python_version >= \"3.10\" and extra == \"mlflow\""}
 
 [package.dependencies]
-protobuf = ">=3.19,<5.0"
+protobuf = ">=5.0,<7.0"
 
 [[package]]
 name = "opentelemetry-sdk"
-version = "1.25.0"
+version = "1.39.1"
 description = "OpenTelemetry Python SDK"
 optional = false
-python-versions = ">=3.8"
+python-versions = ">=3.9"
 groups = ["main", "dev", "proxy-dev"]
 files = [
-    {file = "opentelemetry_sdk-1.25.0-py3-none-any.whl", hash = "sha256:d97ff7ec4b351692e9d5a15af570c693b8715ad78b8aafbec5c7100fe966b4c9"},
-    {file = "opentelemetry_sdk-1.25.0.tar.gz", hash = "sha256:ce7fc319c57707ef5bf8b74fb9f8ebdb8bfafbe11898410e0d2a761d08a98ec7"},
+    {file = "opentelemetry_sdk-1.39.1-py3-none-any.whl", hash = "sha256:4d5482c478513ecb0a5d938dcc61394e647066e0cc2676bee9f3af3f3f45f01c"},
+    {file = "opentelemetry_sdk-1.39.1.tar.gz", hash = "sha256:cf4d4563caf7bff906c9f7967e2be22d0d6b349b908be0d90fb21c8e9c995cc6"},
 ]
 markers = {main = "python_version >= \"3.10\""}
 
 [package.dependencies]
-opentelemetry-api = "1.25.0"
-opentelemetry-semantic-conventions = "0.46b0"
-typing-extensions = ">=3.7.4"
+opentelemetry-api = "1.39.1"
+opentelemetry-semantic-conventions = "0.60b1"
+typing-extensions = ">=4.5.0"
 
 [[package]]
 name = "opentelemetry-semantic-conventions"
-version = "0.46b0"
+version = "0.60b1"
 description = "OpenTelemetry Semantic Conventions"
 optional = false
-python-versions = ">=3.8"
+python-versions = ">=3.9"
 groups = ["main", "dev", "proxy-dev"]
 files = [
-    {file = "opentelemetry_semantic_conventions-0.46b0-py3-none-any.whl", hash = "sha256:6daef4ef9fa51d51855d9f8e0ccd3a1bd59e0e545abe99ac6203804e36ab3e07"},
-    {file = "opentelemetry_semantic_conventions-0.46b0.tar.gz", hash = "sha256:fbc982ecbb6a6e90869b15c1673be90bd18c8a56ff1cffc0864e38e2edffaefa"},
+    {file = "opentelemetry_semantic_conventions-0.60b1-py3-none-any.whl", hash = "sha256:9fa8c8b0c110da289809292b0591220d3a7b53c1526a23021e977d68597893fb"},
+    {file = "opentelemetry_semantic_conventions-0.60b1.tar.gz", hash = "sha256:87c228b5a0669b748c76d76df6c364c369c28f1c465e50f661e39737e84bc953"},
 ]
 markers = {main = "python_version >= \"3.10\""}
 
 [package.dependencies]
-opentelemetry-api = "1.25.0"
+opentelemetry-api = "1.39.1"
+typing-extensions = ">=4.5.0"
 
 [[package]]
 name = "orjson"
@@ -4828,23 +4851,23 @@ testing = ["google-api-core (>=1.31.5)"]
 
 [[package]]
 name = "protobuf"
-version = "4.25.8"
+version = "5.29.5"
 description = ""
 optional = false
 python-versions = ">=3.8"
 groups = ["main", "dev", "proxy-dev"]
 files = [
-    {file = "protobuf-4.25.8-cp310-abi3-win32.whl", hash = "sha256:504435d831565f7cfac9f0714440028907f1975e4bed228e58e72ecfff58a1e0"},
-    {file = "protobuf-4.25.8-cp310-abi3-win_amd64.whl", hash = "sha256:bd551eb1fe1d7e92c1af1d75bdfa572eff1ab0e5bf1736716814cdccdb2360f9"},
-    {file = "protobuf-4.25.8-cp37-abi3-macosx_10_9_universal2.whl", hash = "sha256:ca809b42f4444f144f2115c4c1a747b9a404d590f18f37e9402422033e464e0f"},
-    {file = "protobuf-4.25.8-cp37-abi3-manylinux2014_aarch64.whl", hash = "sha256:9ad7ef62d92baf5a8654fbb88dac7fa5594cfa70fd3440488a5ca3bfc6d795a7"},
-    {file = "protobuf-4.25.8-cp37-abi3-manylinux2014_x86_64.whl", hash = "sha256:83e6e54e93d2b696a92cad6e6efc924f3850f82b52e1563778dfab8b355101b0"},
-    {file = "protobuf-4.25.8-cp38-cp38-win32.whl", hash = "sha256:27d498ffd1f21fb81d987a041c32d07857d1d107909f5134ba3350e1ce80a4af"},
-    {file = "protobuf-4.25.8-cp38-cp38-win_amd64.whl", hash = "sha256:d552c53d0415449c8d17ced5c341caba0d89dbf433698e1436c8fa0aae7808a3"},
-    {file = "protobuf-4.25.8-cp39-cp39-win32.whl", hash = "sha256:077ff8badf2acf8bc474406706ad890466274191a48d0abd3bd6987107c9cde5"},
-    {file = "protobuf-4.25.8-cp39-cp39-win_amd64.whl", hash = "sha256:f4510b93a3bec6eba8fd8f1093e9d7fb0d4a24d1a81377c10c0e5bbfe9e4ed24"},
-    {file = "protobuf-4.25.8-py3-none-any.whl", hash = "sha256:15a0af558aa3b13efef102ae6e4f3efac06f1eea11afb3a57db2901447d9fb59"},
-    {file = "protobuf-4.25.8.tar.gz", hash = "sha256:6135cf8affe1fc6f76cced2641e4ea8d3e59518d1f24ae41ba97bcad82d397cd"},
+    {file = "protobuf-5.29.5-cp310-abi3-win32.whl", hash = "sha256:3f1c6468a2cfd102ff4703976138844f78ebd1fb45f49011afc5139e9e283079"},
+    {file = "protobuf-5.29.5-cp310-abi3-win_amd64.whl", hash = "sha256:3f76e3a3675b4a4d867b52e4a5f5b78a2ef9565549d4037e06cf7b0942b1d3fc"},
+    {file = "protobuf-5.29.5-cp38-abi3-macosx_10_9_universal2.whl", hash = "sha256:e38c5add5a311f2a6eb0340716ef9b039c1dfa428b28f25a7838ac329204a671"},
+    {file = "protobuf-5.29.5-cp38-abi3-manylinux2014_aarch64.whl", hash = "sha256:fa18533a299d7ab6c55a238bf8629311439995f2e7eca5caaff08663606e9015"},
+    {file = "protobuf-5.29.5-cp38-abi3-manylinux2014_x86_64.whl", hash = "sha256:63848923da3325e1bf7e9003d680ce6e14b07e55d0473253a690c3a8b8fd6e61"},
+    {file = "protobuf-5.29.5-cp38-cp38-win32.whl", hash = "sha256:ef91363ad4faba7b25d844ef1ada59ff1604184c0bcd8b39b8a6bef15e1af238"},
+    {file = "protobuf-5.29.5-cp38-cp38-win_amd64.whl", hash = "sha256:7318608d56b6402d2ea7704ff1e1e4597bee46d760e7e4dd42a3d45e24b87f2e"},
+    {file = "protobuf-5.29.5-cp39-cp39-win32.whl", hash = "sha256:6f642dc9a61782fa72b90878af134c5afe1917c89a568cd3476d758d3c3a0736"},
+    {file = "protobuf-5.29.5-cp39-cp39-win_amd64.whl", hash = "sha256:470f3af547ef17847a28e1f47200a1cbf0ba3ff57b7de50d22776607cd2ea353"},
+    {file = "protobuf-5.29.5-py3-none-any.whl", hash = "sha256:6cf42630262c59b2d8de33954443d94b746c952b01434fc58a417fdbd2e84bd5"},
+    {file = "protobuf-5.29.5.tar.gz", hash = "sha256:bc1463bafd4b0929216c35f437a8e28731a2b7fe3d98bb77a600efced5a15c84"},
 ]
 markers = {main = "python_version >= \"3.10\" and (extra == \"mlflow\" or extra == \"extra-proxy\") or extra == \"extra-proxy\""}
 
@@ -7687,7 +7710,7 @@ version = "1.17.3"
 description = "Module for decorators, wrappers and monkey patching."
 optional = false
 python-versions = ">=3.8"
-groups = ["main", "dev", "proxy-dev"]
+groups = ["dev"]
 files = [
     {file = "wrapt-1.17.3-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:88bbae4d40d5a46142e70d58bf664a89b6b4befaea7b2ecc14e03cedb8e06c04"},
     {file = "wrapt-1.17.3-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:e6b13af258d6a9ad602d57d889f83b9d5543acd471eee12eb51f5b01f8eb1bc2"},
@@ -7771,7 +7794,6 @@ files = [
     {file = "wrapt-1.17.3-py3-none-any.whl", hash = "sha256:7171ae35d2c33d326ac19dd8facb1e82e5fd04ef8c6c0e394d7af55a55051c22"},
     {file = "wrapt-1.17.3.tar.gz", hash = "sha256:f66eb08feaa410fe4eebd17f2a2c8e2e46d3476e9f8c783daa8e09e0faa666d0"},
 ]
-markers = {main = "python_version >= \"3.10\""}
 
 [[package]]
 name = "wsproto"
@@ -7972,7 +7994,7 @@ type = ["pytest-mypy"]
 
 [extras]
 caching = ["diskcache"]
-extra-proxy = ["azure-identity", "azure-keyvault-secrets", "google-cloud-iam", "google-cloud-kms", "prisma", "redisvl", "resend"]
+extra-proxy = ["a2a-sdk", "azure-identity", "azure-keyvault-secrets", "google-cloud-iam", "google-cloud-kms", "prisma", "redisvl", "resend"]
 mlflow = ["mlflow"]
 proxy = ["PyJWT", "apscheduler", "azure-identity", "azure-storage-blob", "backoff", "boto3", "cryptography", "fastapi", "fastapi-sso", "gunicorn", "litellm-enterprise", "litellm-proxy-extras", "mcp", "orjson", "polars", "pynacl", "python-multipart", "pyyaml", "rich", "rq", "soundfile", "uvicorn", "uvloop", "websockets"]
 semantic-router = ["semantic-router"]
@@ -7981,4 +8003,4 @@ utils = ["numpydoc"]
 [metadata]
 lock-version = "2.1"
 python-versions = ">=3.9,<4.0"
-content-hash = "ea62b77c662ab9fc486e421c576f0868bcde16d62a24703ee1f4916a0465ffb2"
+content-hash = "82a9400ee6e7a550758024ca3c076c4063c6a22ca9535b125e585b03aca7b2b2"
diff --git a/pyproject.toml b/pyproject.toml
index aa8e6fd97b..c33c8c995f 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -59,6 +59,7 @@ websockets = {version = "^15.0.1", optional = true}
 boto3 = {version = "1.36.0", optional = true}
 redisvl = {version = "^0.4.1", optional = true, markers = "python_version >= '3.9' and python_version < '3.14'"}
 mcp = {version = "^1.21.2", optional = true, python = ">=3.10"}
+a2a-sdk = {version = "^0.3.22", optional = true, python = ">=3.10"}
 litellm-proxy-extras = {version = "0.4.21", optional = true}
 rich = {version = "13.7.1", optional = true}
 litellm-enterprise = {version = "0.1.27", optional = true}
@@ -111,7 +112,8 @@ extra_proxy = [
     "google-cloud-kms",
     "google-cloud-iam",
     "resend",
-    "redisvl"
+    "redisvl",
+    "a2a-sdk"
 ]
 
 utils = [
@@ -147,9 +149,9 @@ types-requests = "*"
 types-setuptools = "*"
 types-redis = "*"
 types-PyYAML = "*"
-opentelemetry-api = "1.25.0"
-opentelemetry-sdk = "1.25.0"
-opentelemetry-exporter-otlp = "1.25.0"
+opentelemetry-api = "^1.28.0"
+opentelemetry-sdk = "^1.28.0"
+opentelemetry-exporter-otlp = "^1.28.0"
 langfuse = "^2.45.0"
 fastapi-offline = "^1.7.3"
 
@@ -157,9 +159,9 @@ fastapi-offline = "^1.7.3"
 prisma = "0.11.0"
 hypercorn = "^0.15.0"
 prometheus-client = "0.20.0"
-opentelemetry-api = "1.25.0"
-opentelemetry-sdk = "1.25.0"
-opentelemetry-exporter-otlp = "1.25.0"
+opentelemetry-api = "^1.28.0"
+opentelemetry-sdk = "^1.28.0"
+opentelemetry-exporter-otlp = "^1.28.0"
 azure-identity = {version = "^1.15.0", python = ">=3.9"}
 
 [build-system]
diff --git a/requirements.txt b/requirements.txt
index 7ec71fb7dc..4fae2591dd 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -16,12 +16,12 @@ prisma==0.11.0 # for db
 nodejs-wheel-binaries==24.12.0 ## required by prisma for migrations, prevents runtime download (updated from nodejs-bin for security fixes)
 mangum==0.17.0 # for aws lambda functions
 pynacl==1.6.2 # for encrypting keys
-google-cloud-aiplatform==1.47.0 # for vertex ai calls
+google-cloud-aiplatform==1.133.0 # for vertex ai calls
 google-cloud-iam==2.19.1 # for GCP IAM Redis authentication
-google-genai==1.22.0
+google-genai==1.37.0
 anthropic[vertex]==0.54.0
 mcp==1.25.0 ; python_version >= "3.10"    # for MCP server
-google-generativeai==0.5.0 # for vertex ai calls
+# google-generativeai removed - deprecated, replaced by google-genai (line 21)
 async_generator==1.10.0 # for async ollama calls
 langfuse==2.59.7 # for langfuse self-hosted logging
 prometheus_client==0.20.0 # for /metrics endpoint on proxy
@@ -37,9 +37,10 @@ azure-ai-contentsafety==1.0.0 # for azure content safety
 azure-identity==1.16.1 ; python_version >= "3.9" # for azure content safety
 azure-keyvault==4.2.0 # for azure KMS integration
 azure-storage-file-datalake==12.20.0 # for azure buck storage logging
-opentelemetry-api==1.25.0
-opentelemetry-sdk==1.25.0
-opentelemetry-exporter-otlp==1.25.0
+opentelemetry-api==1.28.0
+opentelemetry-sdk==1.28.0
+opentelemetry-exporter-otlp==1.28.0
+a2a-sdk>=0.3.22 ; python_version >= "3.10"
 # grpcio: 1.68.0-1.68.1 has reconnect bug (#38290), 1.75+ has Python 3.14 wheels + fix
 grpcio>=1.62.3,!=1.68.*,!=1.69.*,!=1.70.*,!=1.71.0,!=1.71.1,!=1.72.0,!=1.72.1,!=1.73.0; python_version < "3.14"
 grpcio>=1.75.0; python_version >= "3.14"
diff --git a/tests/test_litellm/proxy/agent_endpoints/test_a2a_endpoints.py b/tests/test_litellm/proxy/agent_endpoints/test_a2a_endpoints.py
index 061e27da91..9588c3b55c 100644
--- a/tests/test_litellm/proxy/agent_endpoints/test_a2a_endpoints.py
+++ b/tests/test_litellm/proxy/agent_endpoints/test_a2a_endpoints.py
@@ -49,18 +49,20 @@ async def test_invoke_agent_a2a_adds_litellm_data():
 
     # Mock request
     mock_request = MagicMock()
-    mock_request.json = AsyncMock(return_value={
-        "jsonrpc": "2.0",
-        "id": "test-id",
-        "method": "message/send",
-        "params": {
-            "message": {
-                "role": "user",
-                "parts": [{"kind": "text", "text": "Hello"}],
-                "messageId": "msg-123",
-            }
-        },
-    })
+    mock_request.json = AsyncMock(
+        return_value={
+            "jsonrpc": "2.0",
+            "id": "test-id",
+            "method": "message/send",
+            "params": {
+                "message": {
+                    "role": "user",
+                    "parts": [{"kind": "text", "text": "Hello"}],
+                    "messageId": "msg-123",
+                }
+            },
+        }
+    )
 
     mock_user_api_key_dict = UserAPIKeyAuth(
         api_key="sk-test-key",
@@ -77,40 +79,44 @@ async def test_invoke_agent_a2a_adds_litellm_data():
             SendMessageRequest,
             SendStreamingMessageRequest,
         )
+
         # Real types available - use them
-        use_real_types = True
+        pass
     except ImportError:
         # Real types not available - create realistic mocks
-        use_real_types = False
-        
+        pass
+
         def make_mock_pydantic_class(name):
             """Create a mock class that behaves like a Pydantic model."""
+
             class MockPydanticClass:
                 def __init__(self, **kwargs):
                     self.__dict__.update(kwargs)
                     # Store kwargs for model_dump() if needed
                     self._kwargs = kwargs
-                
+
                 def model_dump(self, mode="json", exclude_none=False):
                     """Mock model_dump method."""
                     result = dict(self._kwargs)
                     if exclude_none:
                         result = {k: v for k, v in result.items() if v is not None}
                     return result
-            
+
             MockPydanticClass.__name__ = name
             return MockPydanticClass
-        
+
         MessageSendParams = make_mock_pydantic_class("MessageSendParams")
         SendMessageRequest = make_mock_pydantic_class("SendMessageRequest")
-        SendStreamingMessageRequest = make_mock_pydantic_class("SendStreamingMessageRequest")
-    
+        SendStreamingMessageRequest = make_mock_pydantic_class(
+            "SendStreamingMessageRequest"
+        )
+
     # Create a mock module for a2a.types
     mock_a2a_types = MagicMock()
     mock_a2a_types.MessageSendParams = MessageSendParams
     mock_a2a_types.SendMessageRequest = SendMessageRequest
     mock_a2a_types.SendStreamingMessageRequest = SendStreamingMessageRequest
-    
+
     # Patch at the source modules
     with patch(
         "litellm.proxy.agent_endpoints.a2a_endpoints._get_agent",
@@ -137,12 +143,15 @@ async def test_invoke_agent_a2a_adds_litellm_data():
     ), patch.dict(
         sys.modules,
         {"a2a": MagicMock(), "a2a.types": mock_a2a_types},
+    ), patch(
+        "litellm.a2a_protocol.main.A2A_SDK_AVAILABLE",
+        True,
     ):
         from litellm.proxy.agent_endpoints.a2a_endpoints import invoke_agent_a2a
 
         mock_fastapi_response = MagicMock()
 
-        result = await invoke_agent_a2a(
+        await invoke_agent_a2a(
             agent_id="test-agent",
             request=mock_request,
             fastapi_response=mock_fastapi_response,

From 896d1a7dad96d1c21b08b65634196d98da6fdfdd Mon Sep 17 00:00:00 2001
From: Sameer Kankute <sameer@berri.ai>
Date: Mon, 19 Jan 2026 18:18:24 +0530
Subject: [PATCH 6/9] Fix Error: Found packages that need verification:

---
 ...model_prices_and_context_window_backup.json | 18 ++++++++++++++++++
 tests/code_coverage_tests/liccheck.ini         |  1 +
 2 files changed, 19 insertions(+)

diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json
index 470d598a25..43f7bde5da 100644
--- a/litellm/model_prices_and_context_window_backup.json
+++ b/litellm/model_prices_and_context_window_backup.json
@@ -7857,6 +7857,24 @@
         "supports_tool_choice": true,
         "supports_vision": true
     },
+    "dall-e-2": {
+        "input_cost_per_image": 0.02,
+        "litellm_provider": "openai",
+        "mode": "image_generation",
+        "supported_endpoints": [
+            "/v1/images/generations",
+            "/v1/images/edits",
+            "/v1/images/variations"
+        ]
+    },
+    "dall-e-3": {
+        "input_cost_per_image": 0.04,
+        "litellm_provider": "openai",
+        "mode": "image_generation",
+        "supported_endpoints": [
+            "/v1/images/generations"
+        ]
+    },
     "deepseek-chat": {
         "cache_read_input_token_cost": 2.8e-08,
         "input_cost_per_token": 2.8e-07,
diff --git a/tests/code_coverage_tests/liccheck.ini b/tests/code_coverage_tests/liccheck.ini
index feb182921d..ea87b56ff3 100644
--- a/tests/code_coverage_tests/liccheck.ini
+++ b/tests/code_coverage_tests/liccheck.ini
@@ -89,6 +89,7 @@ tokenizers: >=0.20.2 # Apache 2.0 License
 jinja2: >=3.1.4 # BSD 3-Clause License
 litellm-proxy-extras: >=0.1.1 # MIT License
 litellm-enterprise: >=0.1.1 # LiteLLM Enterprise License
+a2a-sdk: >=0.3.22 # Apache 2.0 license
 anyio: >=4.5.0 # Unknown license
 httpx-aiohttp: >=0.1.4 # Unknown license
 backoff: >=2.2.1 # Unknown license

From 2e6ce1469e9af016072f6e870f7dae002c0b39ab Mon Sep 17 00:00:00 2001
From: Sameer Kankute <sameer@berri.ai>
Date: Mon, 19 Jan 2026 18:47:28 +0530
Subject: [PATCH 7/9] Fix module import for opentelemetry

---
 litellm/integrations/opentelemetry.py | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/litellm/integrations/opentelemetry.py b/litellm/integrations/opentelemetry.py
index a223925d59..c8fc6e9614 100644
--- a/litellm/integrations/opentelemetry.py
+++ b/litellm/integrations/opentelemetry.py
@@ -986,8 +986,7 @@ class OpenTelemetry(CustomLogger):
         # See: https://github.com/open-telemetry/opentelemetry-python/pull/4676
         # TODO: Refactor to use the proper OTEL Logs API instead of directly creating SDK LogRecords
 
-        from opentelemetry._logs import SeverityNumber, get_logger, get_logger_provider
-        from opentelemetry.sdk._logs import LogRecord as SdkLogRecord
+        from opentelemetry._logs import SeverityNumber, get_logger, get_logger_provider, LogRecord
 
         otel_logger = get_logger(LITELLM_LOGGER_NAME)
 

From 6eb3f579d77e2b4b69f8201635d2ab244016452b Mon Sep 17 00:00:00 2001
From: Sameer Kankute <sameer@berri.ai>
Date: Mon, 19 Jan 2026 18:50:18 +0530
Subject: [PATCH 8/9] Fix module import for opentelemetry

---
 litellm/integrations/opentelemetry.py | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/litellm/integrations/opentelemetry.py b/litellm/integrations/opentelemetry.py
index c8fc6e9614..9fbccac68d 100644
--- a/litellm/integrations/opentelemetry.py
+++ b/litellm/integrations/opentelemetry.py
@@ -986,7 +986,11 @@ class OpenTelemetry(CustomLogger):
         # See: https://github.com/open-telemetry/opentelemetry-python/pull/4676
         # TODO: Refactor to use the proper OTEL Logs API instead of directly creating SDK LogRecords
 
-        from opentelemetry._logs import SeverityNumber, get_logger, get_logger_provider, LogRecord
+        from opentelemetry._logs import SeverityNumber, get_logger, get_logger_provider
+        try:
+            from opentelemetry.sdk._logs import LogRecord as SdkLogRecord  # OTEL < 1.39.0
+        except ImportError:
+            from opentelemetry.sdk._logs._internal import LogRecord as SdkLogRecord  # OTEL >= 1.39.0
 
         otel_logger = get_logger(LITELLM_LOGGER_NAME)
 

From 574391c118c331ade273c4fa21c79ff618358404 Mon Sep 17 00:00:00 2001
From: Sameer Kankute <sameer@berri.ai>
Date: Mon, 19 Jan 2026 18:51:08 +0530
Subject: [PATCH 9/9] Revert "Fix audio cost per second override (#19158)"

This reverts commit 2a0f87bde048e406e6dbf1d5465e77d7e0282bc3.
---
 docs/my-website/docs/sdk_custom_pricing.md |  5 +---
 litellm/llms/openai/cost_calculation.py    | 22 ++++++++------
 tests/test_litellm/test_cost_calculator.py | 34 ----------------------
 3 files changed, 14 insertions(+), 47 deletions(-)

diff --git a/docs/my-website/docs/sdk_custom_pricing.md b/docs/my-website/docs/sdk_custom_pricing.md
index ac956db472..c857711510 100644
--- a/docs/my-website/docs/sdk_custom_pricing.md
+++ b/docs/my-website/docs/sdk_custom_pricing.md
@@ -2,10 +2,7 @@
 
 Register custom pricing for sagemaker completion model. 
 
-For cost per second pricing, register `input_cost_per_second`. If your provider
-charges for audio output duration (e.g., TTS), also set `output_cost_per_second`.
-Values of `0` are treated as not billable, so `output_cost_per_second: 0` will
-not override `input_cost_per_second`.
+For cost per second pricing, you **just** need to register `input_cost_per_second`. 
 
 ```python
 # !pip install boto3 
diff --git a/litellm/llms/openai/cost_calculation.py b/litellm/llms/openai/cost_calculation.py
index 339ce9caed..e5349db3af 100644
--- a/litellm/llms/openai/cost_calculation.py
+++ b/litellm/llms/openai/cost_calculation.py
@@ -105,21 +105,25 @@ def cost_per_second(
     prompt_cost = 0.0
     completion_cost = 0.0
     ## Speech / Audio cost calculation
-    output_cost_per_second = model_info.get("output_cost_per_second")
-    input_cost_per_second = model_info.get("input_cost_per_second")
-
-    if output_cost_per_second is not None and output_cost_per_second > 0:
+    if (
+        "output_cost_per_second" in model_info
+        and model_info["output_cost_per_second"] is not None
+    ):
         verbose_logger.debug(
-            f"For model={model} - output_cost_per_second: {output_cost_per_second}; duration: {duration}"
+            f"For model={model} - output_cost_per_second: {model_info.get('output_cost_per_second')}; duration: {duration}"
         )
         ## COST PER SECOND ##
-        completion_cost = output_cost_per_second * duration
-    if input_cost_per_second is not None and input_cost_per_second > 0:
+        completion_cost = model_info["output_cost_per_second"] * duration
+    elif (
+        "input_cost_per_second" in model_info
+        and model_info["input_cost_per_second"] is not None
+    ):
         verbose_logger.debug(
-            f"For model={model} - input_cost_per_second: {input_cost_per_second}; duration: {duration}"
+            f"For model={model} - input_cost_per_second: {model_info.get('input_cost_per_second')}; duration: {duration}"
         )
         ## COST PER SECOND ##
-        prompt_cost = input_cost_per_second * duration
+        prompt_cost = model_info["input_cost_per_second"] * duration
+        completion_cost = 0.0
 
     return prompt_cost, completion_cost
 
diff --git a/tests/test_litellm/test_cost_calculator.py b/tests/test_litellm/test_cost_calculator.py
index 42bcfca231..4d6599fc1b 100644
--- a/tests/test_litellm/test_cost_calculator.py
+++ b/tests/test_litellm/test_cost_calculator.py
@@ -192,40 +192,6 @@ def test_transcription_cost_falls_back_to_duration():
     assert pytest.approx(cost, rel=1e-6) == expected_cost
 
 
-def test_transcription_cost_prefers_input_when_output_zero(monkeypatch):
-    from litellm import completion_cost
-
-    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
-    litellm.model_cost = litellm.get_model_cost_map(url="")
-
-    model_name = "custom-whisper-input-only"
-    custom_model_info = {
-        "input_cost_per_second": 0.00005,
-        "output_cost_per_second": 0.0,
-        "litellm_provider": "openai",
-        "mode": "audio_transcription",
-        "supported_endpoints": ["/v1/audio/transcriptions"],
-    }
-    monkeypatch.setattr(
-        litellm,
-        "model_cost",
-        {**litellm.model_cost, model_name: custom_model_info},
-    )
-
-    response = TranscriptionResponse(text="demo text")
-    response.duration = 300.0
-
-    cost = completion_cost(
-        completion_response=response,
-        model=model_name,
-        custom_llm_provider="openai",
-        call_type="atranscription",
-    )
-
-    expected_cost = 300.0 * 0.00005
-    assert pytest.approx(cost, rel=1e-6) == expected_cost
-
-
 def test_handle_realtime_stream_cost_calculation():
     from litellm.cost_calculator import RealtimeAPITokenUsageProcessor