Align message-part class names with semconv (*Part suffix)

.github/skills/migrate-from-openinference/SKILL.md

@@ -337,9 +337,9 @@ Drop every match. The mappings:
   `EmbeddingInvocation`, `ToolInvocation`, `WorkflowInvocation`,
   `AgentInvocation`, `Error`, `GenAIInvocation`
 - `opentelemetry.util.genai.types` — `InputMessage`, `OutputMessage`,
-  `Text`, `ToolCallRequest`, `ToolCallResponse`, `Reasoning`,
-  `ServerToolCall`, `ServerToolCallResponse`, `GenericPart`, `Blob`,
-  `File`, `Uri`, `Modality`
+  `TextPart`, `ToolCallRequestPart`, `ToolCallResponsePart`, `ReasoningPart`,
+  `ServerToolCallPart`, `ServerToolCallResponsePart`, `GenericPart`, `BlobPart`,
+  `FilePart`, `UriPart`, `Modality`
 - `opentelemetry.util.genai.completion_hook`
 - `opentelemetry.util.genai.environment_variables`
 
@@ -400,15 +400,15 @@ right (all types from `opentelemetry.util.genai.types` unless noted):
 
 | Source request item | OTel construct |
 |---|---|
-| User / assistant / system text message | `Input/OutputMessage(role=…, parts=[Text(content=…)])` |
-| Assistant message containing a tool/function call | `Message(role="assistant", parts=[ToolCallRequest(name=…, id=…, arguments=…)])` |
-| Tool/function result message | `Message(role="tool", parts=[ToolCallResponse(id=…, response=…)])` |
-| Reasoning / thinking item | `Message(role="assistant", parts=[Reasoning(content=…)])` |
-| Server-side tool call (web_search, file_search, code_interpreter, …) | `Message(parts=[ServerToolCall(name=…, server_tool_call=…, id=…)])` |
-| Server-side tool call result | `Message(parts=[ServerToolCallResponse(server_tool_call_response=…, id=…)])` |
-| Inline image / audio / video bytes | `Blob(mime_type=…, modality="image"\|"audio"\|"video", content=b"…")` |
-| External media URL | `Uri(mime_type=…, modality=…, uri="…")` |
-| File reference (e.g. OpenAI `file_id`) | `File(mime_type=…, modality=…, file_id="file-…")` |
+| User / assistant / system text message | `Input/OutputMessage(role=…, parts=[TextPart(content=…)])` |
+| Assistant message containing a tool/function call | `Message(role="assistant", parts=[ToolCallRequestPart(name=…, id=…, arguments=…)])` |
+| Tool/function result message | `Message(role="tool", parts=[ToolCallResponsePart(id=…, response=…)])` |
+| Reasoning / thinking item | `Message(role="assistant", parts=[ReasoningPart(content=…)])` |
+| Server-side tool call (web_search, file_search, code_interpreter, …) | `Message(parts=[ServerToolCallPart(name=…, server_tool_call=…, id=…)])` |
+| Server-side tool call result | `Message(parts=[ServerToolCallResponsePart(server_tool_call_response=…, id=…)])` |
+| Inline image / audio / video bytes | `BlobPart(mime_type=…, modality="image"\|"audio"\|"video", content=b"…")` |
+| External media URL | `UriPart(mime_type=…, modality=…, uri="…")` |
+| File reference (e.g. OpenAI `file_id`) | `FilePart(mime_type=…, modality=…, file_id="file-…")` |
 | Provider-specific item with no semconv mapping | `GenericPart(value=…)` — never silently drop. Flag those in the review report. |
 
 Output messages mirror the input mapping — `OutputMessage` serializes with

.github/skills/write-conformance-tests/SKILL.md

@@ -58,13 +58,13 @@ instruments: walk its wrappers (the step-6 mapping for a port) for which
 
 | Part `type` | util-genai type | Emitted when the library accepts… |
 |---|---|---|
-| `text` | `Text` | plain text (always) |
-| `tool_call` / `tool_call_response` | `ToolCallRequest` / `ToolCallResponse` | function/tool calling — covered by `tool_calling.py` |
-| `server_tool_call` / `server_tool_call_response` | `ServerToolCall` / `ServerToolCallResponse` | vendor server-side tools (web_search, code_interpreter, …) |
-| `reasoning` | `Reasoning` | reasoning / thinking items |
-| `blob` | `Blob` | inline image/audio/video **bytes** (`modality` distinguishes them) |
-| `uri` | `Uri` | an external media **URL** (`modality`) |
-| `file` | `File` | a **file reference** / id (`modality`) |
+| `text` | `TextPart` | plain text (always) |
+| `tool_call` / `tool_call_response` | `ToolCallRequestPart` / `ToolCallResponsePart` | function/tool calling — covered by `tool_calling.py` |
+| `server_tool_call` / `server_tool_call_response` | `ServerToolCallPart` / `ServerToolCallResponsePart` | vendor server-side tools (web_search, code_interpreter, …) |
+| `reasoning` | `ReasoningPart` | reasoning / thinking items |
+| `blob` | `BlobPart` | inline image/audio/video **bytes** (`modality` distinguishes them) |
+| `uri` | `UriPart` | an external media **URL** (`modality`) |
+| `file` | `FilePart` | a **file reference** / id (`modality`) |
 | `generic` | `GenericPart` | a provider item with no semconv mapping — flag, don't drop |
 
 Group by shared turn/cassette — typically one `multimodal.py` for the

instrumentation/opentelemetry-instrumentation-genai-agno/.changelog/365.changed

@@ -0,0 +1 @@
+Align message-part class names with semconv (*Part suffix). Rename Text/Blob/File/Uri/Reasoning/ToolCallRequest/ToolCallResponse/ServerToolCall/ServerToolCallResponse to their *Part forms to match semantic-conventions-genai models.

instrumentation/opentelemetry-instrumentation-genai-agno/pyproject.toml

@@ -28,7 +28,7 @@ dependencies = [
   "opentelemetry-api ~= 1.43",
   "opentelemetry-instrumentation >= 0.64b0, <1",
   "opentelemetry-semantic-conventions >= 0.64b0, <1",
-  "opentelemetry-util-genai >= 1.0b0, <2",
+  "opentelemetry-util-genai >= 1.1b0.dev, <2",
 ]
 
 [project.optional-dependencies]

instrumentation/opentelemetry-instrumentation-genai-agno/src/opentelemetry/instrumentation/genai/agno/patch.py

@@ -24,7 +24,7 @@
 from opentelemetry.util.genai.types import (
     InputMessage,
     OutputMessage,
-    Text,
+    TextPart,
 )
 
 logger = logging.getLogger(__name__)
@@ -151,7 +151,9 @@ def _set_invocation_input(
         if input_val is not None:
             content_str = _extract_input_content(input_val)
             invocation.input_messages = [
-                InputMessage(role="user", parts=[Text(content=content_str)])
+                InputMessage(
+                    role="user", parts=[TextPart(content=content_str)]
+                )
             ]
 
 
@@ -165,7 +167,7 @@ def _set_invocation_output(
         invocation.output_messages = [
             OutputMessage(
                 role="assistant",
-                parts=[Text(content=output_str)],
+                parts=[TextPart(content=output_str)],
                 finish_reason="stop",
             )
         ]

instrumentation/opentelemetry-instrumentation-genai-agno/tests/requirements.oldest.txt

@@ -21,5 +21,5 @@
 # OpenTelemetry SDK and test utilities come transitively from opentelemetry-test-util-genai, which
 # every oldest env installs. Pin here only test-only deps that nothing else already provides.
 #
-# There is currently nothing to pin: agno has no test-only dependency that isn't already
-# provided by a declared bound or by the shared test fixtures.
+# Drop this once opentelemetry-util-genai 1.1b0 is published.
+-e util/opentelemetry-util-genai

instrumentation/opentelemetry-instrumentation-genai-anthropic/.changelog/365.changed

@@ -0,0 +1 @@
+Align message-part class names with semconv (*Part suffix). Rename Text/Blob/File/Uri/Reasoning/ToolCallRequest/ToolCallResponse/ServerToolCall/ServerToolCallResponse to their *Part forms to match semantic-conventions-genai models.

instrumentation/opentelemetry-instrumentation-genai-anthropic/src/opentelemetry/instrumentation/genai/anthropic/utils.py

@@ -24,12 +24,12 @@
 )
 
 from opentelemetry.util.genai.types import (
-    Blob,
+    BlobPart,
     MessagePart,
-    Reasoning,
-    Text,
-    ToolCallRequest,
-    ToolCallResponse,
+    ReasoningPart,
+    TextPart,
+    ToolCallRequestPart,
+    ToolCallResponsePart,
 )
 
 if TYPE_CHECKING:
@@ -72,8 +72,8 @@ def _decode_base64(data: str) -> bytes | None:
         return None
 
 
-def _extract_base64_blob(source: object, modality: str) -> Blob | None:
-    """Extract a Blob from a base64-encoded source dict."""
+def _extract_base64_blob(source: object, modality: str) -> BlobPart | None:
+    """Extract a BlobPart from a base64-encoded source dict."""
     if not isinstance(source, dict):
         return None
     # source is a TypedDict (e.g. Base64ImageSourceParam) narrowed to dict;
@@ -85,7 +85,7 @@ def _extract_base64_blob(source: object, modality: str) -> Blob | None:
     if decoded is None:
         return None
     media_type: object = source.get("media_type")  # type: ignore[reportUnknownMemberType]
-    return Blob(
+    return BlobPart(
         mime_type=media_type if isinstance(media_type, str) else None,
         modality=modality,
         content=decoded,
@@ -100,25 +100,27 @@ def _convert_dict_block_to_part(
 
     if block_type == "text":
         text = block.get("text")
-        return Text(content=str(text) if text is not None else "")
+        return TextPart(content=str(text) if text is not None else "")
 
     if block_type == "tool_use":
         inp = block.get("input")
-        return ToolCallRequest(
+        return ToolCallRequestPart(
             arguments=inp if isinstance(inp, dict) else None,
             name=str(block.get("name", "")),
             id=str(block.get("id", "")),
         )
 
     if block_type == "tool_result":
-        return ToolCallResponse(
+        return ToolCallResponsePart(
             response=block.get("content"),
             id=str(block.get("tool_use_id", "")),
         )
 
     if block_type in ("thinking", "redacted_thinking"):
         thinking = block.get("thinking") or block.get("data")
-        return Reasoning(content=str(thinking) if thinking is not None else "")
+        return ReasoningPart(
+            content=str(thinking) if thinking is not None else ""
+        )
 
     if block_type in ("image", "audio", "video", "document", "file"):
         return _extract_base64_blob(block.get("source"), str(block_type))
@@ -131,21 +133,21 @@ def _convert_content_block_to_part(
 ) -> MessagePart | None:
     """Convert an Anthropic content block to a MessagePart."""
     if isinstance(block, TextBlock):
-        return Text(content=block.text)
+        return TextPart(content=block.text)
 
     if isinstance(block, (ToolUseBlock, ServerToolUseBlock)):
-        return ToolCallRequest(
+        return ToolCallRequestPart(
             arguments=block.input, name=block.name, id=block.id
         )
 
     if isinstance(block, (ThinkingBlock, RedactedThinkingBlock)):
         content = (
             block.thinking if isinstance(block, ThinkingBlock) else block.data
         )
-        return Reasoning(content=content)
+        return ReasoningPart(content=content)
 
     if isinstance(block, WebSearchToolResultBlock):
-        return ToolCallResponse(
+        return ToolCallResponsePart(
             response=block.model_dump().get("content"),
             id=block.tool_use_id,
         )
@@ -161,7 +163,7 @@ def convert_content_to_parts(
     if content is None:
         return []
     if isinstance(content, str):
-        return [Text(content=content)]
+        return [TextPart(content=content)]
     parts: list[MessagePart] = []
     for item in content:
         part = _convert_content_block_to_part(item)
@@ -209,7 +211,7 @@ def update_stream_block_state(
 
 def stream_block_state_to_part(state: StreamBlockState) -> MessagePart | None:
     if state.type == "text":
-        return Text(content=state.text)
+        return TextPart(content=state.text)
 
     if state.type == "tool_use":
         arguments: str | dict[str, object] | None = state.tool_input
@@ -218,13 +220,13 @@ def stream_block_state_to_part(state: StreamBlockState) -> MessagePart | None:
                 arguments = json.loads(state.input_json)
             except ValueError:
                 arguments = state.input_json
-        return ToolCallRequest(
+        return ToolCallRequestPart(
             arguments=arguments,
             name=state.tool_name,
             id=state.tool_id,
         )
 
     if state.type in ("thinking", "redacted_thinking"):
-        return Reasoning(content=state.thinking)
+        return ReasoningPart(content=state.thinking)
 
     return None

instrumentation/opentelemetry-instrumentation-genai-langchain/.changelog/365.changed

@@ -0,0 +1 @@
+Align message-part class names with semconv (*Part suffix). Rename Text/Blob/File/Uri/Reasoning/ToolCallRequest/ToolCallResponse/ServerToolCall/ServerToolCallResponse to their *Part forms to match semantic-conventions-genai models.

instrumentation/opentelemetry-instrumentation-genai-langchain/src/opentelemetry/instrumentation/genai/langchain/callback_handler.py

@@ -42,8 +42,8 @@
 from opentelemetry.util.genai.types import (
     MessagePart,
     OutputMessage,
-    Text,
-    ToolCallRequest,
+    TextPart,
+    ToolCallRequestPart,
 )
 
 SUPPORTED_RAPI_RESPONSE_HEADERS = ("x-ms-served-model",)
@@ -374,9 +374,9 @@ def on_llm_end(
                         )
 
                     if finish_reason in ("tool_calls", "tool_use"):
-                        tool_calls: list[ToolCallRequest] = []
+                        tool_calls: list[ToolCallRequestPart] = []
                         for tool_call in chat_generation.message.tool_calls:
-                            tool_call_request = ToolCallRequest(
+                            tool_call_request = ToolCallRequestPart(
                                 name=tool_call["name"],
                                 id=tool_call["id"],
                                 arguments=tool_call["args"],
@@ -402,7 +402,7 @@ def on_llm_end(
                         )
                     else:
                         parts = [
-                            Text(
+                            TextPart(
                                 content=chat_generation.message.content,
                                 type="text",
                             )

instrumentation/opentelemetry-instrumentation-genai-langchain/src/opentelemetry/instrumentation/genai/langchain/utils.py

@@ -23,10 +23,10 @@
     InputMessage,
     MessagePart,
     OutputMessage,
-    Reasoning,
-    Text,
-    ToolCallRequest,
-    ToolCallResponse,
+    ReasoningPart,
+    TextPart,
+    ToolCallRequestPart,
+    ToolCallResponsePart,
     ToolDefinition,
 )
 
@@ -81,41 +81,41 @@ def _content_to_parts(
     """Convert a LangChain message ``content`` payload into ``MessagePart`` s.
 
     Content may be a plain string or a list of provider-specific block dicts
-    (e.g. Anthropic structured content). We extract :class:`Text` and
-    :class:`Reasoning` parts; ``tool_use`` blocks are intentionally ignored
+    (e.g. Anthropic structured content). We extract :class:`TextPart` and
+    :class:`ReasoningPart` parts; ``tool_use`` blocks are intentionally ignored
     here because LangChain consolidates them into ``message.tool_calls`` which
     is read separately.
     """
     parts: list[MessagePart] = []
     if isinstance(content, str):
         if content:
-            parts.append(Text(content=content))
+            parts.append(TextPart(content=content))
         return parts
     for item in content:
         if isinstance(item, str):
             if item:
-                parts.append(Text(content=item))
+                parts.append(TextPart(content=item))
             continue
         block_type = item.get("type")
         if block_type == "text":
             text_value = item.get("text")
             if isinstance(text_value, str) and text_value:
-                parts.append(Text(content=text_value))
+                parts.append(TextPart(content=text_value))
         elif block_type in ("thinking", "reasoning"):
             reasoning_value = (
                 item.get("thinking")
                 or item.get("reasoning")
                 or item.get("text")
             )
             if isinstance(reasoning_value, str) and reasoning_value:
-                parts.append(Reasoning(content=reasoning_value))
+                parts.append(ReasoningPart(content=reasoning_value))
     return parts
 
 
 def _legacy_function_call_request(
     message: AIMessage,
-) -> ToolCallRequest | None:
-    """Extract a legacy OpenAI ``function_call`` as a :class:`ToolCallRequest`.
+) -> ToolCallRequestPart | None:
+    """Extract a legacy OpenAI ``function_call`` as a :class:`ToolCallRequestPart`.
 
     Pre-tools OpenAI models return a single call under
     ``additional_kwargs['function_call']`` (``{"name", "arguments"}``) rather
@@ -136,14 +136,14 @@ def _legacy_function_call_request(
             arguments = json.loads(raw_arguments)
         except (json.JSONDecodeError, ValueError):
             arguments = raw_arguments
-    return ToolCallRequest(arguments=arguments, name=name, id=None)
+    return ToolCallRequestPart(arguments=arguments, name=name, id=None)
 
 
 def _ai_message_parts(message: AIMessage) -> list[MessagePart]:
     """Build :class:`MessagePart` s for an :class:`AIMessage`.
 
     Includes any text/reasoning content followed by a
-    :class:`ToolCallRequest` for each entry in ``message.tool_calls``, plus a
+    :class:`ToolCallRequestPart` for each entry in ``message.tool_calls``, plus a
     legacy ``additional_kwargs['function_call']`` when present.
     """
     parts: list[MessagePart] = _content_to_parts(message.content)
@@ -152,7 +152,7 @@ def _ai_message_parts(message: AIMessage) -> list[MessagePart]:
         if not name:
             continue
         parts.append(
-            ToolCallRequest(
+            ToolCallRequestPart(
                 arguments=call["args"],
                 name=name,
                 id=call["id"],
@@ -169,7 +169,7 @@ def _tool_message_parts(message: ToolMessage) -> list[MessagePart]:
     """Build :class:`MessagePart` s for a :class:`ToolMessage` (tool result)."""
     tool_call_id = getattr(message, "tool_call_id", None)
     return [
-        ToolCallResponse(
+        ToolCallResponsePart(
             response=message.content,
             id=tool_call_id if isinstance(tool_call_id, str) else None,
         )
@@ -294,7 +294,7 @@ def make_input_message(data: Any) -> list[InputMessage]:
 
     When no ``messages`` key exists (common in LangGraph state dicts), the
     remaining state fields are serialized as JSON and emitted as a single
-    user-role :class:`Text` part.
+    user-role :class:`TextPart` part.
     """
     if not isinstance(data, dict):
         return []
@@ -318,7 +318,7 @@ def make_input_message(data: Any) -> list[InputMessage]:
     if input_data:
         serialized = serialize(input_data)
         if serialized:
-            return [InputMessage(role="user", parts=[Text(serialized)])]
+            return [InputMessage(role="user", parts=[TextPart(serialized)])]
     return []