Promote: int-agentic → main (production)

agentic_ai/agents/agent_framework/STATE_MANAGEMENT.md

@@ -408,7 +408,7 @@ class SessionMemoryCheckpointStorage(CheckpointStorage):
 
 2. **Team agent**  
    - `MagenticOrchestratorExecutor` automatically snapshots each participant’s `chat_history`.
-   - Checkpoints include outstanding tool requests (`RequestInfoExecutor` state).
+   - Checkpoints include outstanding tool requests and plan-review state managed by the orchestrator via `WorkflowEvent(type="request_info")`.
    - Ensure all custom executors implement `restore_state` to interpret snapshots after schema changes.
 
 3. **Human-in-the-loop**  

agentic_ai/agents/agent_framework/multi_agent/MAGENTIC_README.md

@@ -128,7 +128,7 @@ When stalling, the manager calls `replan()`, which uses [`ORCHESTRATOR_TASK_LEDG
 1. Registers **participants** (`participants()`), each either a `BaseAgent` or another `Executor`.
 2. Configures the **manager** (`with_standard_manager()`).
 3. Optionally enables **plan review**, **checkpointing**, and **unified callbacks**.
-4. Builds a [`Workflow`](reference/agent-framework/python/packages/main/agent_framework/_workflow/_workflow.py) graph: orchestrator (start node), optional `RequestInfoExecutor` for plan review, and bidirectional edges to each agent executor.
+4. Builds a [`Workflow`](reference/agent-framework/python/packages/main/agent_framework/_workflow/_workflow.py) graph: orchestrator (start node) with optional plan review enabled via `enable_plan_review=True`, and bidirectional edges to each agent executor. Plan review is handled internally by the orchestrator, which emits `WorkflowEvent(type="request_info")` events containing `MagenticPlanReviewRequest` data.
 
 ### 4.2 Workflow Context & State
 

agentic_ai/agents/agent_framework/multi_agent/handoff_multi_domain_agent.py

@@ -286,10 +286,10 @@ async def _setup(self) -> None:
         has_api_key = bool(self.azure_openai_key)
         has_credential = bool(self.azure_credential)
 
-        if not all([self.azure_deployment, self.azure_openai_endpoint, self.api_version]):
+        if not all([self.azure_deployment, self.azure_openai_endpoint]):
             raise RuntimeError(
                 "Azure OpenAI configuration is incomplete. Ensure "
-                "AZURE_OPENAI_CHAT_DEPLOYMENT, AZURE_OPENAI_ENDPOINT, and AZURE_OPENAI_API_VERSION are set."
+                "AZURE_OPENAI_CHAT_DEPLOYMENT and AZURE_OPENAI_ENDPOINT are set."
             )
         if not has_api_key and not has_credential:
             raise RuntimeError(
@@ -311,15 +311,13 @@ async def _setup(self) -> None:
                 api_key=self.azure_openai_key,
                 model=self.azure_deployment,
                 azure_endpoint=self.azure_openai_endpoint,
-                api_version=self.api_version,
             )
             logger.info("[HANDOFF] Using API key authentication for Azure OpenAI")
         else:
             chat_client = OpenAIChatClient(
                 credential=self.azure_credential,
                 model=self.azure_deployment,
                 azure_endpoint=self.azure_openai_endpoint,
-                api_version=self.api_version,
             )
             logger.info("[HANDOFF] Using managed identity authentication for Azure OpenAI")
 
@@ -340,7 +338,7 @@ async def _setup(self) -> None:
                 description=cfg["description"],
                 instructions=cfg["instructions"],
                 tools=domain_tools,
-                default_options=ChatOptions(model_id=self.openai_model_name),
+                default_options=ChatOptions(model=self.openai_model_name),
                 require_per_service_call_history_persistence=True,
             )
             await agent.__aenter__()

agentic_ai/agents/agent_framework/multi_agent/magentic_group.py

@@ -295,10 +295,15 @@ async def chat_async(self, prompt: str) -> str:
         return cleaned_answer
 
     def _validate_configuration(self) -> None:
-        if not all([self.azure_openai_key, self.azure_deployment, self.azure_openai_endpoint, self.api_version]):
+        if not all([self.azure_deployment, self.azure_openai_endpoint]):
             raise RuntimeError(
-                "Azure OpenAI configuration is incomplete. Ensure AZURE_OPENAI_API_KEY, "
-                "AZURE_OPENAI_CHAT_DEPLOYMENT, AZURE_OPENAI_ENDPOINT, and AZURE_OPENAI_API_VERSION are set."
+                "Azure OpenAI configuration is incomplete. Ensure AZURE_OPENAI_CHAT_DEPLOYMENT "
+                "and AZURE_OPENAI_ENDPOINT are set."
+            )
+        if not self.azure_openai_key and not getattr(self, "azure_credential", None):
+            raise RuntimeError(
+                "Azure OpenAI authentication is not configured. Set AZURE_OPENAI_API_KEY "
+                "or provide a managed-identity credential."
             )
 
     def _build_headers(self) -> Dict[str, str]:
@@ -376,15 +381,13 @@ def _build_chat_client(self) -> OpenAIChatClient:
                 api_key=self.azure_openai_key,
                 model=self.azure_deployment,
                 azure_endpoint=self.azure_openai_endpoint,
-                api_version=self.api_version,
             )
         elif self.azure_credential:
             logger.info("[AgentFramework-Magentic] Using managed identity authentication for Azure OpenAI")
             return OpenAIChatClient(
                 credential=self.azure_credential,
                 model=self.azure_deployment,
                 azure_endpoint=self.azure_openai_endpoint,
-                api_version=self.api_version,
             )
         else:
             raise RuntimeError(
@@ -579,7 +582,7 @@ async def _create_participants(
             agent_kwargs: Dict[str, Any] = {
                 **defaults,
                 "client": participant_client,
-                "default_options": ChatOptions(model_id=self.openai_model_name),
+                "default_options": ChatOptions(model=self.openai_model_name),
             }
 
             # Apply tool filtering for this participant's domain

agentic_ai/agents/agent_framework/multi_agent/reflection_agent.py

@@ -80,7 +80,7 @@ async def _setup_agents(self) -> None:
             return
 
         # Validate configuration
-        if not all([self.azure_deployment, self.azure_openai_endpoint, self.api_version]):
+        if not all([self.azure_deployment, self.azure_openai_endpoint]):
             raise RuntimeError("Azure OpenAI configuration incomplete.")
 
         if not self.azure_openai_key and not self.azure_credential:
@@ -90,7 +90,6 @@ async def _setup_agents(self) -> None:
         client_kwargs = {
             "model": self.azure_deployment,
             "azure_endpoint": self.azure_openai_endpoint,
-            "api_version": self.api_version,
         }
         if self.azure_openai_key:
             client_kwargs["api_key"] = self.azure_openai_key
@@ -108,15 +107,15 @@ async def _setup_agents(self) -> None:
             name="PrimaryAgent",
             instructions=PRIMARY_AGENT_INSTRUCTIONS,
             tools=tools,
-            default_options=ChatOptions(model_id=self.openai_model_name),
+            default_options=ChatOptions(model=self.openai_model_name),
         )
 
         self._reviewer = FrameworkAgent(
             client=chat_client,
             name="Reviewer",
             instructions=REVIEWER_INSTRUCTIONS,
             tools=tools,
-            default_options=ChatOptions(model_id=self.openai_model_name),
+            default_options=ChatOptions(model=self.openai_model_name),
         )
 
         # Initialize agents

agentic_ai/agents/agent_framework/single_agent.py

@@ -38,10 +38,10 @@ async def _setup_single_agent(self) -> None:
         has_api_key = bool(self.azure_openai_key)
         has_credential = bool(self.azure_credential)
 
-        if not all([self.azure_deployment, self.azure_openai_endpoint, self.api_version]):
+        if not all([self.azure_deployment, self.azure_openai_endpoint]):
             raise RuntimeError(
                 "Azure OpenAI configuration is incomplete. Ensure "
-                "AZURE_OPENAI_CHAT_DEPLOYMENT, AZURE_OPENAI_ENDPOINT, and AZURE_OPENAI_API_VERSION are set."
+                "AZURE_OPENAI_CHAT_DEPLOYMENT and AZURE_OPENAI_ENDPOINT are set."
             )
 
         if not has_api_key and not has_credential:
@@ -59,15 +59,13 @@ async def _setup_single_agent(self) -> None:
                 api_key=self.azure_openai_key,
                 model=self.azure_deployment,
                 azure_endpoint=self.azure_openai_endpoint,
-                api_version=self.api_version,
             )
             logger.info("[AgentFramework] Using API key authentication for Azure OpenAI")
         else:
             chat_client = OpenAIChatClient(
                 credential=self.azure_credential,
                 model=self.azure_deployment,
                 azure_endpoint=self.azure_openai_endpoint,
-                api_version=self.api_version,
             )
             logger.info("[AgentFramework] Using managed identity authentication for Azure OpenAI")
 
@@ -85,7 +83,7 @@ async def _setup_single_agent(self) -> None:
             name="ai_assistant",
             instructions=instructions,
             tools=tools,
-            default_options=ChatOptions(model_id=self.openai_model_name),
+            default_options=ChatOptions(model=self.openai_model_name),
         )
 
         try:

infra/terraform/_aca-be.tf

@@ -85,11 +85,6 @@ resource "azurerm_container_app" "backend" {
         value = local.openai_endpoint
       }
 
-      env {
-        name  = "AZURE_OPENAI_API_VERSION"
-        value = var.openai_api_version
-      }
-
       env {
         name  = "AZURE_OPENAI_EMBEDDING_DEPLOYMENT"
         value = var.openai_embedding_deployment_name

infra/terraform/dev.tfvars

@@ -14,7 +14,6 @@ create_openai_deployment = true
 openai_deployment_name   = "gpt-5.2-chat"
 openai_model_name        = "gpt-5.2-chat"
 openai_model_version     = "2025-12-11"
-openai_api_version       ="2025-04-01-preview"
 openai_deployment_capacity = 200  # 200k tokens/minute - needed for agent evaluation
 
 # OpenAI embedding deployment configuration

infra/terraform/prod.tfvars

@@ -14,7 +14,6 @@ create_openai_deployment = true
 openai_deployment_name   = "gpt-5.2-chat"
 openai_model_name        = "gpt-5.2-chat"
 openai_model_version     = "2025-12-11"
-openai_api_version       = "2025-04-01-preview"
 openai_deployment_capacity = 200  # 200k tokens/minute
 
 # OpenAI embedding deployment configuration

infra/terraform/variables.tf

@@ -41,11 +41,6 @@ variable "openai_model_version" {
   type        = string
   default     = "2025-12-11"
 }
-variable "openai_api_version" {
-  description = "OpenAI API version"
-  type        = string
-  default     = "2025-04-01-preview"
-}
 variable "openai_deployment_capacity" {
   description = "Capacity (TPM in thousands) for OpenAI deployment"
   type        = number

tests/test_agent_framework_1_2_1_regression.py

@@ -149,11 +149,25 @@ def test_framework_agent_constructor_signature(self):
         # description was wired through in 1.2.x and is consumed by HandoffBuilder
         assert 'description' in params
 
-    def test_chat_options_model_id(self):
-        """ChatOptions accepts model_id for specifying the model."""
+    def test_chat_options_model(self):
+        """ChatOptions exposes ``model`` (not ``model_id``) in 1.2.x.
+
+        Regression for the CI failure where agents constructed
+        ``ChatOptions(model_id=...)``. ``ChatOptions`` is a ``TypedDict`` with
+        ``total=False``, so unknown keys are silently accepted at construction
+        time and only blow up later when forwarded as kwargs to the OpenAI
+        ``responses.create`` call (``TypeError: unexpected keyword argument
+        'model_id'``). Assert against ``__annotations__`` so the test fails
+        loudly if the field name changes again.
+        """
         from agent_framework import ChatOptions
-        opts = ChatOptions(model_id="gpt-4o")
-        assert opts["model_id"] == "gpt-4o"
+        opts = ChatOptions(model="gpt-4o")
+        assert opts["model"] == "gpt-4o"
+        assert "model" in ChatOptions.__annotations__
+        assert "model_id" not in ChatOptions.__annotations__, (
+            "ChatOptions field is `model` in agent-framework 1.2.x; "
+            "passing model_id= silently constructs a bad dict that fails downstream."
+        )
 
     def test_openai_client_constructor_supports_azure_kwargs(self):
         """OpenAIChatClient accepts model, api_key, credential, azure_endpoint, api_version."""