feat: add ResponseFormat support to call_model_input_filter (#3563)

src/agents/run_config.py

@@ -28,7 +28,7 @@
     from .sandbox.session.sandbox_client import BaseSandboxClient
     from .sandbox.session.sandbox_session_state import SandboxSessionState
     from .sandbox.snapshot import SnapshotBase, SnapshotSpec
-
+from openai.types.chat.completion_create_params import ResponseFormat
 
 DEFAULT_MAX_TURNS = 10
 DEFAULT_MAX_MANIFEST_ENTRY_CONCURRENCY = 4
@@ -50,6 +50,7 @@ class ModelInputData:
 
     input: list[TResponseInputItem]
     instructions: str | None
+    response_format: ResponseFormat | None = None
 
 
 @dataclass

src/agents/run_internal/run_loop.py

@@ -194,6 +194,7 @@
     resolve_interrupted_turn,
     run_final_output_hooks,
 )
+from openai.types.chat.completion_create_params import ResponseFormat
 
 __all__ = [
     "extract_tool_call_id",
@@ -1366,13 +1367,27 @@ def _tool_search_fingerprint(raw_item: Any) -> str:
             streamed_result._model_input_items,
             reasoning_item_id_policy,
         )
+    model_settings = get_model_settings(execution_agent, run_config)
+    model_settings = maybe_reset_tool_choice(public_agent, tool_use_tracker, model_settings)
+
+    current_response_format: ResponseFormat | None = None
+    if output_schema and hasattr(output_schema, "is_plain_text") and not output_schema.is_plain_text():
+        current_response_format = cast(ResponseFormat, {
+            "type": "json_schema",
+            "json_schema": {
+                "name": "final_output",
+                "strict": output_schema.is_strict_json_schema(),
+                "schema": output_schema.json_schema(),
+            },
+        })
 
     filtered = await maybe_filter_model_input(
         agent=public_agent,
         run_config=run_config,
         context_wrapper=context_wrapper,
         input_items=input,
         system_instructions=system_prompt,
+        response_format=current_response_format,
     )
     if isinstance(filtered.input, list):
         filtered.input = deduplicate_input_items_preferring_latest(filtered.input)
@@ -1815,19 +1830,33 @@ async def get_new_response(
     """Call the model and return the raw response, handling retries and hooks."""
     public_agent = bindings.public_agent
     execution_agent = bindings.execution_agent
+
+    model = get_model(execution_agent, run_config)
+    model_settings = get_model_settings(execution_agent, run_config)
+    model_settings = maybe_reset_tool_choice(public_agent, tool_use_tracker, model_settings)
+
+    current_response_format: ResponseFormat | None = None
+    if output_schema and hasattr(output_schema, "is_plain_text") and not output_schema.is_plain_text():
+        current_response_format = cast(ResponseFormat, {
+            "type": "json_schema",
+            "json_schema": {
+                "name": "final_output",
+                "strict": output_schema.is_strict_json_schema(),
+                "schema": output_schema.json_schema(),
+            },
+        })
+
     filtered = await maybe_filter_model_input(
         agent=public_agent,
         run_config=run_config,
         context_wrapper=context_wrapper,
         input_items=input,
         system_instructions=system_prompt,
+        response_format=current_response_format,
     )
     if isinstance(filtered.input, list):
         filtered.input = deduplicate_input_items_preferring_latest(filtered.input)
 
-    model = get_model(execution_agent, run_config)
-    model_settings = get_model_settings(execution_agent, run_config)
-    model_settings = maybe_reset_tool_choice(public_agent, tool_use_tracker, model_settings)
 
     if server_conversation_tracker is not None:
         server_conversation_tracker.mark_input_as_sent(filtered.input)

src/agents/run_internal/turn_preparation.py

@@ -18,6 +18,7 @@
 from ..tool import Tool
 from ..tracing import SpanError
 from ..util import _error_tracing
+from openai.types.chat import ResponseFormat
 
 __all__ = [
     "validate_run_hooks",
@@ -55,18 +56,24 @@ async def maybe_filter_model_input(
     context_wrapper: RunContextWrapper[TContext],
     input_items: list[TResponseInputItem],
     system_instructions: str | None,
+    response_format: ResponseFormat | None = None,
 ) -> ModelInputData:
     """Apply optional call_model_input_filter to modify model input."""
     effective_instructions = system_instructions
     effective_input: list[TResponseInputItem] = input_items
 
     if run_config.call_model_input_filter is None:
-        return ModelInputData(input=effective_input, instructions=effective_instructions)
+        return ModelInputData(
+            input=effective_input,
+            instructions=effective_instructions,
+            response_format=response_format,
+        )
 
     try:
         model_input = ModelInputData(
             input=effective_input.copy(),
             instructions=effective_instructions,
+            response_format=response_format,
         )
         filter_payload: CallModelData[TContext] = CallModelData(
             model_data=model_input,