agent-framework-devui 1.0.0b260821__tar.gz → 1.0.0b260910__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (26) hide show
  1. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/PKG-INFO +4 -4
  2. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/_conversations.py +176 -35
  3. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/_executor.py +126 -65
  4. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/_mapper.py +181 -81
  5. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/_openai/__init__.py +9 -9
  6. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/_server.py +3 -27
  7. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/_utils.py +29 -0
  8. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/models/__init__.py +98 -94
  9. agent_framework_devui-1.0.0b260910/agent_framework_devui/ui/assets/index.js +193 -0
  10. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/pyproject.toml +5 -5
  11. agent_framework_devui-1.0.0b260821/agent_framework_devui/ui/assets/index.js +0 -199
  12. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/LICENSE +0 -0
  13. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/README.md +0 -0
  14. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/__init__.py +0 -0
  15. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/_cli.py +0 -0
  16. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/_deployment.py +0 -0
  17. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/_discovery.py +0 -0
  18. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/_feature_usage.py +0 -0
  19. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/_openai/_executor.py +0 -0
  20. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/_session.py +0 -0
  21. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/_tracing.py +0 -0
  22. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/models/_discovery_models.py +0 -0
  23. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/models/_openai_custom.py +0 -0
  24. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/ui/agentframework.svg +0 -0
  25. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/ui/assets/index.css +0 -0
  26. {agent_framework_devui-1.0.0b260821 → agent_framework_devui-1.0.0b260910}/agent_framework_devui/ui/index.html +0 -0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: agent-framework-devui
3
- Version: 1.0.0b260821
3
+ Version: 1.0.0b260910
4
4
  Summary: Debug UI for Microsoft Agent Framework with OpenAI-compatible API server.
5
5
  Author-email: Microsoft <af-support@microsoft.com>
6
6
  Requires-Python: >=3.10
@@ -16,10 +16,10 @@ Classifier: Programming Language :: Python :: 3.13
16
16
  Classifier: Programming Language :: Python :: 3.14
17
17
  Classifier: Typing :: Typed
18
18
  License-File: LICENSE
19
- Requires-Dist: agent-framework-core>=1.15.0,<2
20
- Requires-Dist: openai>=2.45.0,<3
19
+ Requires-Dist: agent-framework-core>=1.17.0,<2
20
+ Requires-Dist: openai>=2.45.0,<4
21
21
  Requires-Dist: opentelemetry-sdk>=1.39.0,<2
22
- Requires-Dist: fastapi>=0.115.0,<0.138.1
22
+ Requires-Dist: fastapi>=0.115.0,<0.142.0
23
23
  Requires-Dist: uvicorn[standard]>=0.30.0,<1
24
24
  Requires-Dist: pytest==9.1.1 ; extra == "all"
25
25
  Requires-Dist: watchdog==6.0.0 ; extra == "all"
@@ -14,7 +14,7 @@ from abc import ABC, abstractmethod
14
14
  from collections.abc import MutableSequence
15
15
  from typing import Any, Literal, cast
16
16
 
17
- from agent_framework import AgentSession, Message
17
+ from agent_framework import AgentSession, Content, Message
18
18
  from agent_framework._workflows._checkpoint import InMemoryCheckpointStorage, WorkflowCheckpoint
19
19
  from openai.types.conversations import Conversation, ConversationDeletedResource
20
20
  from openai.types.conversations.conversation_item import ConversationItem
@@ -26,8 +26,14 @@ from openai.types.responses import (
26
26
  ResponseFunctionToolCallOutputItem,
27
27
  ResponseInputFile,
28
28
  ResponseInputImage,
29
+ ResponseOutputRefusal,
29
30
  )
30
31
 
32
+ from ._utils import infer_media_type
33
+
34
+ _MODEL_OUTPUT_KIND_KEY = "model_output_kind"
35
+ _MODEL_OUTPUT_REFUSAL = "refusal"
36
+
31
37
  # Type alias for OpenAI Message role literals
32
38
  MessageRole = Literal["unknown", "user", "assistant", "system", "critic", "discriminator", "developer", "tool"]
33
39
 
@@ -304,17 +310,23 @@ class InMemoryConversationStore(ConversationStore):
304
310
  # Convert items to Messages and add to storage
305
311
  chat_messages: list[Message] = []
306
312
  for item in items:
307
- # Simple conversion - assume text content for now
308
313
  role = item.get("role", "user")
309
314
  content = item.get("content", [])
310
- first_content = cast(
311
- dict[str, Any],
312
- content[0] if content and isinstance(content, list) and isinstance(content[0], dict) else {},
313
- )
314
- text_obj = first_content.get("text", "")
315
- text = text_obj if isinstance(text_obj, str) else str(text_obj)
315
+ contents: list[Content] = []
316
316
 
317
- chat_msg = Message(role=role, contents=[text])
317
+ if isinstance(content, str):
318
+ contents.append(Content.from_text(text=content))
319
+ elif isinstance(content, list):
320
+ for content_item in cast(list[object], content):
321
+ if isinstance(content_item, dict):
322
+ agent_content = self._to_agent_content(cast(dict[str, Any], content_item))
323
+ if agent_content is not None:
324
+ contents.append(agent_content)
325
+
326
+ if not contents:
327
+ raise ValueError("Conversation message did not contain any supported content")
328
+
329
+ chat_msg = Message(role=role, contents=contents)
318
330
  chat_messages.append(chat_msg)
319
331
 
320
332
  # Add messages to internal storage
@@ -328,9 +340,9 @@ class InMemoryConversationStore(ConversationStore):
328
340
  # Convert Message contents to OpenAI TextContent format
329
341
  message_content: MutableSequence[OpenAIContent] = []
330
342
  for content_item in msg.contents:
331
- if content_item.type == "text":
332
- # Extract text from TextContent object
333
- message_content.append(TextContent(type="text", text=content_item.text or ""))
343
+ openai_content = self._to_openai_content(content_item)
344
+ if openai_content is not None:
345
+ message_content.append(openai_content)
334
346
 
335
347
  # Create Message object (concrete type from ConversationItem union)
336
348
  message = OpenAIMessage(
@@ -384,35 +396,16 @@ class InMemoryConversationStore(ConversationStore):
384
396
  # Process each content item in the message
385
397
  # A single Message may produce multiple ConversationItems
386
398
  # (e.g., a message with both text and a function call)
387
- message_contents: list[TextContent | ResponseInputImage | ResponseInputFile] = []
399
+ message_contents: list[TextContent | ResponseOutputRefusal | ResponseInputImage | ResponseInputFile] = []
388
400
  function_calls: list[ResponseFunctionToolCallItem] = []
389
401
  function_results: list[ResponseFunctionToolCallOutputItem] = []
390
402
 
391
403
  for content in msg.contents:
392
404
  content_type = getattr(content, "type", None)
405
+ mapped_content = self._to_openai_content(content)
393
406
 
394
- if content_type == "text":
395
- # Text content for Message
396
- text_value = getattr(content, "text", "")
397
- message_contents.append(TextContent(type="text", text=text_value))
398
-
399
- elif content_type == "data":
400
- # Data content (images, files, PDFs)
401
- uri = getattr(content, "uri", "")
402
- media_type = getattr(content, "media_type", None)
403
-
404
- if media_type and media_type.startswith("image/"):
405
- # Convert to ResponseInputImage
406
- message_contents.append(ResponseInputImage(type="input_image", image_url=uri, detail="auto"))
407
- else:
408
- # Convert to ResponseInputFile
409
- # Extract filename from URI if possible
410
- filename = None
411
- if media_type == "application/pdf":
412
- filename = "document.pdf"
413
-
414
- message_contents.append(ResponseInputFile(type="input_file", file_url=uri, filename=filename))
415
-
407
+ if mapped_content is not None:
408
+ message_contents.append(mapped_content)
416
409
  elif content_type == "function_call":
417
410
  # Function call - create separate ConversationItem
418
411
  call_id = getattr(content, "call_id", None)
@@ -532,6 +525,154 @@ class InMemoryConversationStore(ConversationStore):
532
525
 
533
526
  return paginated_items, has_more
534
527
 
528
+ @staticmethod
529
+ def _to_agent_content(content: dict[str, Any]) -> Content | None:
530
+ """Convert one supported OpenAI conversation message part."""
531
+ content_type = content.get("type")
532
+ if content_type in ("text", "input_text", "output_text"):
533
+ text = content.get("text")
534
+ return Content.from_text(text=text) if isinstance(text, str) else None
535
+
536
+ if content_type == "refusal":
537
+ refusal = content.get("refusal")
538
+ if not isinstance(refusal, str):
539
+ return None
540
+ return Content.from_text(
541
+ text=refusal,
542
+ additional_properties={_MODEL_OUTPUT_KIND_KEY: _MODEL_OUTPUT_REFUSAL},
543
+ )
544
+
545
+ if content_type == "input_image":
546
+ detail = content.get("detail", "auto")
547
+ image_properties = {
548
+ "openai_content_type": "input_image",
549
+ "detail": detail if isinstance(detail, str) else "auto",
550
+ }
551
+ image_url = content.get("image_url")
552
+ if isinstance(image_url, str) and image_url:
553
+ return Content.from_uri(
554
+ uri=image_url,
555
+ media_type=infer_media_type(uri=image_url, default="image/png"),
556
+ additional_properties=image_properties,
557
+ )
558
+ file_id = content.get("file_id")
559
+ if isinstance(file_id, str) and file_id:
560
+ return Content.from_hosted_file(
561
+ file_id=file_id,
562
+ additional_properties=image_properties,
563
+ )
564
+ return None
565
+
566
+ if content_type == "input_file":
567
+ filename = content.get("filename")
568
+ filename = filename if isinstance(filename, str) else ""
569
+ detail = content.get("detail")
570
+ file_properties: dict[str, Any] = {"openai_content_type": "input_file"}
571
+ if filename:
572
+ file_properties["filename"] = filename
573
+ if isinstance(detail, str):
574
+ file_properties["detail"] = detail
575
+
576
+ file_data = content.get("file_data")
577
+ if isinstance(file_data, str) and file_data:
578
+ media_type = infer_media_type(
579
+ filename=filename,
580
+ uri=file_data,
581
+ default="application/octet-stream",
582
+ )
583
+ uri = file_data if file_data.startswith("data:") else f"data:{media_type};base64,{file_data}"
584
+ return Content.from_uri(
585
+ uri=uri,
586
+ media_type=media_type,
587
+ additional_properties=file_properties,
588
+ )
589
+
590
+ file_url = content.get("file_url")
591
+ if isinstance(file_url, str) and file_url:
592
+ return Content.from_uri(
593
+ uri=file_url,
594
+ media_type=infer_media_type(
595
+ filename=filename,
596
+ uri=file_url,
597
+ default="application/octet-stream",
598
+ ),
599
+ additional_properties=file_properties,
600
+ )
601
+
602
+ file_id = content.get("file_id")
603
+ if isinstance(file_id, str) and file_id:
604
+ return Content.from_hosted_file(
605
+ file_id=file_id,
606
+ media_type=infer_media_type(filename=filename, default="application/octet-stream"),
607
+ name=filename or None,
608
+ additional_properties=file_properties,
609
+ )
610
+
611
+ return None
612
+
613
+ @staticmethod
614
+ def _to_openai_content(
615
+ content: Content,
616
+ ) -> TextContent | ResponseOutputRefusal | ResponseInputImage | ResponseInputFile | None:
617
+ """Convert one supported Agent Framework message part."""
618
+ content_type = content.type
619
+ if content_type == "text":
620
+ if content.additional_properties.get(_MODEL_OUTPUT_KIND_KEY) == _MODEL_OUTPUT_REFUSAL:
621
+ return ResponseOutputRefusal(type="refusal", refusal=content.text or "")
622
+ return TextContent(type="text", text=content.text or "")
623
+
624
+ additional_properties = content.additional_properties or {}
625
+ openai_content_type = additional_properties.get("openai_content_type")
626
+ is_image = openai_content_type == "input_image" or (
627
+ openai_content_type is None and bool(content.media_type and content.media_type.startswith("image/"))
628
+ )
629
+
630
+ if content_type in ("data", "uri"):
631
+ if is_image:
632
+ detail = additional_properties.get("detail", "auto")
633
+ if detail not in ("low", "high", "auto", "original"):
634
+ detail = "auto"
635
+ return ResponseInputImage(
636
+ type="input_image",
637
+ image_url=content.uri,
638
+ detail=detail,
639
+ )
640
+
641
+ filename = additional_properties.get("filename")
642
+ detail = additional_properties.get("detail")
643
+ if detail not in ("low", "high", "auto"):
644
+ detail = None
645
+ return ResponseInputFile(
646
+ type="input_file",
647
+ file_url=content.uri,
648
+ filename=filename if isinstance(filename, str) else None,
649
+ detail=detail,
650
+ )
651
+
652
+ if content_type == "hosted_file":
653
+ if is_image:
654
+ detail = additional_properties.get("detail", "auto")
655
+ if detail not in ("low", "high", "auto", "original"):
656
+ detail = "auto"
657
+ return ResponseInputImage(
658
+ type="input_image",
659
+ file_id=content.file_id,
660
+ detail=detail,
661
+ )
662
+
663
+ filename = content.name or additional_properties.get("filename")
664
+ detail = additional_properties.get("detail")
665
+ if detail not in ("low", "high", "auto"):
666
+ detail = None
667
+ return ResponseInputFile(
668
+ type="input_file",
669
+ file_id=content.file_id,
670
+ filename=filename if isinstance(filename, str) else None,
671
+ detail=detail,
672
+ )
673
+
674
+ return None
675
+
535
676
  async def get_item(self, conversation_id: str, item_id: str) -> ConversationItem | None:
536
677
  """Get a specific conversation item by ID.
537
678
 
@@ -15,11 +15,14 @@ from ._conversations import ConversationStore, InMemoryConversationStore
15
15
  from ._discovery import EntityDiscovery
16
16
  from ._mapper import MessageMapper
17
17
  from ._tracing import capture_traces
18
+ from ._utils import infer_media_type
18
19
  from .models import AgentFrameworkRequest, OpenAIResponse
19
20
  from .models._discovery_models import EntityInfo
20
21
 
21
22
  logger = logging.getLogger(__name__)
22
23
 
24
+ _OPENAI_MESSAGE_ROLES = {"user", "assistant", "system", "developer"}
25
+
23
26
 
24
27
  def _get_event_type(event: Any) -> str | None:
25
28
  """Safely get the type of an event, handling both objects and dicts."""
@@ -35,6 +38,10 @@ class EntityNotFoundError(Exception):
35
38
  pass
36
39
 
37
40
 
41
+ class InputConversionError(ValueError):
42
+ """Raised when OpenAI input contains no supported Agent Framework content."""
43
+
44
+
38
45
  class AgentFrameworkExecutor:
39
46
  """Executor for Agent Framework entities - agents and workflows."""
40
47
 
@@ -128,7 +135,9 @@ class AgentFrameworkExecutor:
128
135
  request_id = event.get("request_id")
129
136
  fc = event.get("function_call", {})
130
137
  if isinstance(request_id, str) and request_id:
138
+ occurrence_id = fc.get("occurrence_id")
131
139
  self._pending_approvals[request_id] = {
140
+ "id": occurrence_id if isinstance(occurrence_id, str) and occurrence_id else request_id,
132
141
  "call_id": fc.get("id", ""),
133
142
  "name": fc.get("name", ""),
134
143
  "arguments": fc.get("arguments", {}),
@@ -229,6 +238,7 @@ class AgentFrameworkExecutor:
229
238
  OpenAI response stream events
230
239
  """
231
240
  try:
241
+ events: list[Any] = []
232
242
  entity_id = request.get_entity_id()
233
243
  if not entity_id:
234
244
  logger.error("No entity_id specified in request")
@@ -249,8 +259,13 @@ class AgentFrameworkExecutor:
249
259
  and cast(dict[str, Any], event).get("type") == "response.function_approval.requested"
250
260
  ):
251
261
  self._track_approval_request(cast(dict[str, Any], event))
262
+ events.append(event)
263
+ if _get_event_type(event) == "response.failed":
264
+ continue
252
265
  yield event
253
266
 
267
+ yield await self.message_mapper.finalize_stream(events, request)
268
+
254
269
  except Exception as e:
255
270
  logger.exception(f"Error in streaming execution: {e}")
256
271
  # Could yield error event here
@@ -267,7 +282,14 @@ class AgentFrameworkExecutor:
267
282
  # Collect all streaming events
268
283
  events = [event async for event in self.execute_streaming(request)]
269
284
 
270
- # Aggregate into final response
285
+ for event in reversed(events):
286
+ if _get_event_type(event) not in ("response.completed", "response.failed"):
287
+ continue
288
+ response = getattr(event, "response", None)
289
+ if isinstance(response, OpenAIResponse):
290
+ return response
291
+
292
+ # Preserve the existing empty-response fallback when execution produced no events.
271
293
  return await self.message_mapper.aggregate_to_response(events, request)
272
294
 
273
295
  async def execute_entity(self, entity_id: str, request: AgentFrameworkRequest) -> AsyncGenerator[Any]:
@@ -592,6 +614,10 @@ class AgentFrameworkExecutor:
592
614
  # The workflow is already paused by ctx.request_info() in the framework
593
615
  # DevUI should continue yielding events even during HIL pause
594
616
 
617
+ except InputConversionError as e:
618
+ from .models._openai_custom import AgentFailedEvent
619
+
620
+ yield AgentFailedEvent(error=e)
595
621
  except Exception as e:
596
622
  logger.error(f"Error in workflow execution: {e}")
597
623
  yield {"type": "error", "message": f"Workflow execution error: {e!s}"}
@@ -610,7 +636,7 @@ class AgentFrameworkExecutor:
610
636
  """
611
637
  # Import Agent Framework types
612
638
  try:
613
- from agent_framework import Message, Role
639
+ from agent_framework import Message
614
640
  except ImportError:
615
641
  # Fallback to string extraction if Agent Framework not available
616
642
  return self._extract_user_message_fallback(input_data)
@@ -622,13 +648,13 @@ class AgentFrameworkExecutor:
622
648
  # Handle OpenAI ResponseInputParam (List[ResponseInputItemParam])
623
649
  if isinstance(input_data, list):
624
650
  input_items: Any = cast(Any, input_data)
625
- return self._convert_openai_input_to_chat_message(input_items, Message, Role)
651
+ return self._convert_openai_input_to_chat_message(input_items, Message)
626
652
 
627
653
  # Fallback for other formats
628
654
  return self._extract_user_message_fallback(input_data)
629
655
 
630
- def _convert_openai_input_to_chat_message(self, input_items: list[Any], Message: Any, Role: Any) -> Any:
631
- """Convert OpenAI ResponseInputParam to Agent Framework Message.
656
+ def _convert_openai_input_to_chat_message(self, input_items: list[Any], Message: Any) -> Any:
657
+ """Convert OpenAI ResponseInputParam to Agent Framework messages.
632
658
 
633
659
  Processes text, images, files, and other content types from OpenAI format
634
660
  to Agent Framework Message with appropriate content objects.
@@ -636,12 +662,12 @@ class AgentFrameworkExecutor:
636
662
  Args:
637
663
  input_items: List of OpenAI ResponseInputItemParam objects (dicts or objects)
638
664
  Message: Message class for creating chat messages
639
- Role: Role enum for message roles
640
665
 
641
666
  Returns:
642
- Message with converted content
667
+ One Message, or a list of Messages when input contains multiple message items
643
668
  """
644
- contents: list[Content] = []
669
+ messages: list[Any] = []
670
+ approval_request_ids: set[str] = set()
645
671
 
646
672
  # Process each input item
647
673
  for item in input_items:
@@ -649,9 +675,15 @@ class AgentFrameworkExecutor:
649
675
  if isinstance(item, dict):
650
676
  item_dict = cast(dict[str, Any], item)
651
677
  item_type = item_dict.get("type")
652
- if item_type == "message":
678
+ is_easy_message = item_type is None and "role" in item_dict and "content" in item_dict
679
+ if item_type == "message" or is_easy_message:
680
+ role_value = item_dict.get("role", "user")
681
+ if not isinstance(role_value, str) or role_value not in _OPENAI_MESSAGE_ROLES:
682
+ raise InputConversionError(f"Unsupported OpenAI input message role: {role_value!r}")
683
+
653
684
  # Extract content from OpenAI message
654
685
  message_content = item_dict.get("content", [])
686
+ contents: list[Content] = []
655
687
 
656
688
  # Handle both string content and list content
657
689
  if isinstance(message_content, str):
@@ -664,78 +696,97 @@ class AgentFrameworkExecutor:
664
696
  content_dict = cast(dict[str, Any], content_item)
665
697
  content_type = content_dict.get("type")
666
698
 
667
- if content_type == "input_text":
699
+ if content_type in ("input_text", "output_text", "text"):
668
700
  text = content_dict.get("text", "")
669
701
  if isinstance(text, str):
670
702
  contents.append(Content.from_text(text=text))
671
703
 
672
704
  elif content_type == "input_image":
673
705
  image_url = content_dict.get("image_url", "")
706
+ file_id = content_dict.get("file_id")
707
+ detail = content_dict.get("detail", "auto")
708
+ image_properties = {
709
+ "openai_content_type": "input_image",
710
+ "detail": detail if isinstance(detail, str) else "auto",
711
+ }
674
712
  if isinstance(image_url, str) and image_url:
675
- # Extract media type from data URI if possible
676
- # Parse media type from data URL, fallback to image/png
677
- if image_url.startswith("data:"):
678
- try:
679
- # Extract media type from data:image/jpeg;base64,... format
680
- media_type = image_url.split(";")[0].split(":")[1]
681
- except (IndexError, AttributeError):
682
- logger.warning(
683
- f"Failed to parse media type from data URL: {image_url[:30]}..."
684
- )
685
- media_type = "image/png"
686
- else:
687
- media_type = "image/png"
688
- contents.append(Content.from_uri(uri=image_url, media_type=media_type))
713
+ media_type = infer_media_type(uri=image_url, default="image/png")
714
+ contents.append(
715
+ Content.from_uri(
716
+ uri=image_url,
717
+ media_type=media_type,
718
+ additional_properties=image_properties,
719
+ )
720
+ )
721
+ elif isinstance(file_id, str) and file_id:
722
+ contents.append(
723
+ Content.from_hosted_file(
724
+ file_id=file_id,
725
+ additional_properties=image_properties,
726
+ )
727
+ )
689
728
 
690
729
  elif content_type == "input_file":
691
730
  # Handle file input
692
731
  file_data = content_dict.get("file_data")
732
+ file_id = content_dict.get("file_id")
693
733
  file_url = content_dict.get("file_url")
694
734
  filename = content_dict.get("filename", "")
735
+ detail = content_dict.get("detail")
695
736
 
696
737
  if not isinstance(filename, str):
697
738
  filename = ""
698
739
 
699
- # Determine media type from filename
700
- media_type = "application/octet-stream" # default
701
- if filename:
702
- if filename.lower().endswith(".pdf"):
703
- media_type = "application/pdf"
704
- elif filename.lower().endswith((".png", ".jpg", ".jpeg", ".gif")):
705
- media_type = f"image/{filename.split('.')[-1].lower()}"
706
- elif filename.lower().endswith((
707
- ".wav",
708
- ".mp3",
709
- ".m4a",
710
- ".ogg",
711
- ".flac",
712
- ".aac",
713
- )):
714
- ext = filename.split(".")[-1].lower()
715
- # Normalize extensions to match audio MIME types
716
- media_type = "audio/mp4" if ext == "m4a" else f"audio/{ext}"
717
-
718
740
  # Use file_data or file_url
719
741
  # Include filename in additional_properties for OpenAI/Azure file handling
720
- additional_props: dict[str, Any] | None = (
721
- {"filename": filename} if filename else None
722
- )
742
+ file_properties: dict[str, Any] = {"openai_content_type": "input_file"}
743
+ if filename:
744
+ file_properties["filename"] = filename
745
+ if isinstance(detail, str):
746
+ file_properties["detail"] = detail
747
+
723
748
  if isinstance(file_data, str) and file_data:
724
- # Assume file_data is base64, create data URI
725
- data_uri = f"data:{media_type};base64,{file_data}"
749
+ media_type = infer_media_type(
750
+ filename=filename,
751
+ uri=file_data,
752
+ default="application/octet-stream",
753
+ )
754
+ data_uri = (
755
+ file_data
756
+ if file_data.startswith("data:")
757
+ else f"data:{media_type};base64,{file_data}"
758
+ )
726
759
  contents.append(
727
760
  Content.from_uri(
728
761
  uri=data_uri,
729
762
  media_type=media_type,
730
- additional_properties=additional_props,
763
+ additional_properties=file_properties,
731
764
  )
732
765
  )
733
766
  elif isinstance(file_url, str) and file_url:
767
+ media_type = infer_media_type(
768
+ filename=filename,
769
+ uri=file_url,
770
+ default="application/octet-stream",
771
+ )
734
772
  contents.append(
735
773
  Content.from_uri(
736
774
  uri=file_url,
737
775
  media_type=media_type,
738
- additional_properties=additional_props,
776
+ additional_properties=file_properties,
777
+ )
778
+ )
779
+ elif isinstance(file_id, str) and file_id:
780
+ media_type = infer_media_type(
781
+ filename=filename,
782
+ default="application/octet-stream",
783
+ )
784
+ contents.append(
785
+ Content.from_hosted_file(
786
+ file_id=file_id,
787
+ media_type=media_type,
788
+ name=filename or None,
789
+ additional_properties=file_properties,
739
790
  )
740
791
  )
741
792
 
@@ -752,7 +803,11 @@ class AgentFrameworkExecutor:
752
803
 
753
804
  # Only accept responses that match a request we issued.
754
805
  # Always use the server-stored function_call data.
755
- stored_fc = self._pending_approvals.pop(request_id, None)
806
+ stored_fc = (
807
+ None
808
+ if request_id in approval_request_ids
809
+ else self._pending_approvals.get(request_id)
810
+ )
756
811
  if stored_fc is None:
757
812
  logger.warning(
758
813
  "Rejected function_approval_response with unknown "
@@ -777,6 +832,7 @@ class AgentFrameworkExecutor:
777
832
  name=stored_fc["name"],
778
833
  arguments=stored_fc["arguments"],
779
834
  )
835
+ function_call.id = stored_fc.get("id", request_id)
780
836
 
781
837
  # Create approval response using server-validated data
782
838
  approval_response = Content.from_function_approval_response(
@@ -786,6 +842,7 @@ class AgentFrameworkExecutor:
786
842
  additional_properties=approval_additional_props,
787
843
  )
788
844
  contents.append(approval_response)
845
+ approval_request_ids.add(request_id)
789
846
  logger.info(
790
847
  "Validated FunctionApprovalResponseContent: id=%s, "
791
848
  "approved=%s, function=%s, policy_violation=%s",
@@ -801,24 +858,23 @@ class AgentFrameworkExecutor:
801
858
  except Exception as e:
802
859
  logger.error(f"Failed to process FunctionApprovalResponseContent: {e}")
803
860
 
861
+ if not contents:
862
+ raise InputConversionError("OpenAI input message did not contain any supported message content")
863
+
864
+ messages.append(Message(role=role_value, contents=contents))
865
+
804
866
  # Handle other OpenAI input item types as needed
805
867
  # (tool calls, function results, etc.)
806
868
 
807
- # If no contents found, create a simple text message
808
- if not contents:
809
- contents.append(Content.from_text(text=""))
869
+ if not messages:
870
+ raise InputConversionError("OpenAI input did not contain any supported message content")
810
871
 
811
- chat_message = Message(role="user", contents=contents)
872
+ for request_id in approval_request_ids:
873
+ self._pending_approvals.pop(request_id, None)
812
874
 
813
- logger.info(f"Created Message with {len(contents)} contents:")
814
- for idx, content in enumerate(contents):
815
- content_type = content.__class__.__name__
816
- if hasattr(content, "media_type"):
817
- logger.info(f" [{idx}] {content_type} - media_type: {content.media_type}")
818
- else:
819
- logger.info(f" [{idx}] {content_type}")
875
+ logger.info("Created %d Message object(s) from OpenAI input", len(messages))
820
876
 
821
- return chat_message
877
+ return messages[0] if len(messages) == 1 else messages
822
878
 
823
879
  def _extract_user_message_fallback(self, input_data: Any) -> str:
824
880
  """Fallback method to extract user message as string.
@@ -857,8 +913,11 @@ class AgentFrameworkExecutor:
857
913
  first_item = input_data_items[0]
858
914
  if not isinstance(first_item, dict):
859
915
  return False
860
- first_type = cast(dict[str, Any], first_item).get("type")
861
- return isinstance(first_type, str) and first_type == "message"
916
+ first_item_dict = cast(dict[str, Any], first_item)
917
+ first_type = first_item_dict.get("type")
918
+ return first_type == "message" or (
919
+ first_type is None and "role" in first_item_dict and "content" in first_item_dict
920
+ )
862
921
 
863
922
  async def _parse_workflow_input(self, workflow: Any, raw_input: Any) -> Any:
864
923
  """Parse input based on workflow's expected input type.
@@ -893,6 +952,8 @@ class AgentFrameworkExecutor:
893
952
  # Handle string input
894
953
  return self._parse_raw_workflow_input(workflow, str(raw_input))
895
954
 
955
+ except InputConversionError:
956
+ raise
896
957
  except Exception as e:
897
958
  logger.warning(f"Error parsing workflow input: {e}")
898
959
  return cast(Any, raw_input)