openai-agents 0.0.11__py3-none-any.whl → 0.0.13__py3-none-any.whl

agents/mcp/server.py

@@ -3,6 +3,7 @@ from __future__ import annotations
 import abc
 import asyncio
 from contextlib import AbstractAsyncContextManager, AsyncExitStack
+from datetime import timedelta
 from pathlib import Path
 from typing import Any, Literal
 
@@ -54,7 +55,7 @@ class MCPServer(abc.ABC):
 class _MCPServerWithClientSession(MCPServer, abc.ABC):
     """Base class for MCP servers that use a `ClientSession` to communicate with the server."""
 
-    def __init__(self, cache_tools_list: bool):
+    def __init__(self, cache_tools_list: bool, client_session_timeout_seconds: float | None):
         """
         Args:
             cache_tools_list: Whether to cache the tools list. If `True`, the tools list will be
@@ -63,12 +64,16 @@ class _MCPServerWithClientSession(MCPServer, abc.ABC):
             by calling `invalidate_tools_cache()`. You should set this to `True` if you know the
             server will not change its tools list, because it can drastically improve latency
             (by avoiding a round-trip to the server every time).
+
+            client_session_timeout_seconds: the read timeout passed to the MCP ClientSession.
         """
         self.session: ClientSession | None = None
         self.exit_stack: AsyncExitStack = AsyncExitStack()
         self._cleanup_lock: asyncio.Lock = asyncio.Lock()
         self.cache_tools_list = cache_tools_list
 
+        self.client_session_timeout_seconds = client_session_timeout_seconds
+
         # The cache is always dirty at startup, so that we fetch tools at least once
         self._cache_dirty = True
         self._tools_list: list[MCPTool] | None = None
@@ -101,7 +106,15 @@ class _MCPServerWithClientSession(MCPServer, abc.ABC):
         try:
             transport = await self.exit_stack.enter_async_context(self.create_streams())
             read, write = transport
-            session = await self.exit_stack.enter_async_context(ClientSession(read, write))
+            session = await self.exit_stack.enter_async_context(
+                ClientSession(
+                    read,
+                    write,
+                    timedelta(seconds=self.client_session_timeout_seconds)
+                    if self.client_session_timeout_seconds
+                    else None,
+                )
+            )
             await session.initialize()
             self.session = session
         except Exception as e:
@@ -137,9 +150,10 @@ class _MCPServerWithClientSession(MCPServer, abc.ABC):
         async with self._cleanup_lock:
             try:
                 await self.exit_stack.aclose()
-                self.session = None
             except Exception as e:
                 logger.error(f"Error cleaning up server: {e}")
+            finally:
+                self.session = None
 
 
 class MCPServerStdioParams(TypedDict):
@@ -182,6 +196,7 @@ class MCPServerStdio(_MCPServerWithClientSession):
         params: MCPServerStdioParams,
         cache_tools_list: bool = False,
         name: str | None = None,
+        client_session_timeout_seconds: float | None = 5,
     ):
         """Create a new MCP server based on the stdio transport.
 
@@ -198,8 +213,9 @@ class MCPServerStdio(_MCPServerWithClientSession):
                 improve latency (by avoiding a round-trip to the server every time).
             name: A readable name for the server. If not provided, we'll create one from the
                 command.
+            client_session_timeout_seconds: the read timeout passed to the MCP ClientSession.
         """
-        super().__init__(cache_tools_list)
+        super().__init__(cache_tools_list, client_session_timeout_seconds)
 
         self.params = StdioServerParameters(
             command=params["command"],
@@ -256,6 +272,7 @@ class MCPServerSse(_MCPServerWithClientSession):
         params: MCPServerSseParams,
         cache_tools_list: bool = False,
         name: str | None = None,
+        client_session_timeout_seconds: float | None = 5,
     ):
         """Create a new MCP server based on the HTTP with SSE transport.
 
@@ -273,8 +290,10 @@ class MCPServerSse(_MCPServerWithClientSession):
 
             name: A readable name for the server. If not provided, we'll create one from the
                 URL.
+
+            client_session_timeout_seconds: the read timeout passed to the MCP ClientSession.
         """
-        super().__init__(cache_tools_list)
+        super().__init__(cache_tools_list, client_session_timeout_seconds)
 
         self.params = params
         self._name = name or f"sse: {self.params['url']}"

agents/model_settings.py

@@ -1,10 +1,12 @@
 from __future__ import annotations
 
+import dataclasses
 from dataclasses import dataclass, fields, replace
-from typing import Literal
+from typing import Any, Literal
 
-from openai._types import Body, Query
+from openai._types import Body, Headers, Query
 from openai.types.shared import Reasoning
+from pydantic import BaseModel
 
 
 @dataclass
@@ -67,6 +69,10 @@ class ModelSettings:
     """Additional body fields to provide with the request.
     Defaults to None if not provided."""
 
+    extra_headers: Headers | None = None
+    """Additional headers to provide with the request.
+    Defaults to None if not provided."""
+
     def resolve(self, override: ModelSettings | None) -> ModelSettings:
         """Produce a new ModelSettings by overlaying any non-None values from the
         override on top of this instance."""
@@ -79,3 +85,16 @@ class ModelSettings:
             if getattr(override, field.name) is not None
         }
         return replace(self, **changes)
+
+    def to_json_dict(self) -> dict[str, Any]:
+        dataclass_dict = dataclasses.asdict(self)
+
+        json_dict: dict[str, Any] = {}
+
+        for field_name, value in dataclass_dict.items():
+            if isinstance(value, BaseModel):
+                json_dict[field_name] = value.model_dump(mode="json")
+            else:
+                json_dict[field_name] = value
+
+        return json_dict

agents/models/chatcmpl_converter.py

@@ -0,0 +1,466 @@
+from __future__ import annotations
+
+import json
+from collections.abc import Iterable
+from typing import Any, Literal, cast
+
+from openai import NOT_GIVEN, NotGiven
+from openai.types.chat import (
+    ChatCompletionAssistantMessageParam,
+    ChatCompletionContentPartImageParam,
+    ChatCompletionContentPartParam,
+    ChatCompletionContentPartTextParam,
+    ChatCompletionDeveloperMessageParam,
+    ChatCompletionMessage,
+    ChatCompletionMessageParam,
+    ChatCompletionMessageToolCallParam,
+    ChatCompletionSystemMessageParam,
+    ChatCompletionToolChoiceOptionParam,
+    ChatCompletionToolMessageParam,
+    ChatCompletionUserMessageParam,
+)
+from openai.types.chat.chat_completion_tool_param import ChatCompletionToolParam
+from openai.types.chat.completion_create_params import ResponseFormat
+from openai.types.responses import (
+    EasyInputMessageParam,
+    ResponseFileSearchToolCallParam,
+    ResponseFunctionToolCall,
+    ResponseFunctionToolCallParam,
+    ResponseInputContentParam,
+    ResponseInputImageParam,
+    ResponseInputTextParam,
+    ResponseOutputMessage,
+    ResponseOutputMessageParam,
+    ResponseOutputRefusal,
+    ResponseOutputText,
+)
+from openai.types.responses.response_input_param import FunctionCallOutput, ItemReference, Message
+
+from ..agent_output import AgentOutputSchemaBase
+from ..exceptions import AgentsException, UserError
+from ..handoffs import Handoff
+from ..items import TResponseInputItem, TResponseOutputItem
+from ..tool import FunctionTool, Tool
+from .fake_id import FAKE_RESPONSES_ID
+
+
+class Converter:
+    @classmethod
+    def convert_tool_choice(
+        cls, tool_choice: Literal["auto", "required", "none"] | str | None
+    ) -> ChatCompletionToolChoiceOptionParam | NotGiven:
+        if tool_choice is None:
+            return NOT_GIVEN
+        elif tool_choice == "auto":
+            return "auto"
+        elif tool_choice == "required":
+            return "required"
+        elif tool_choice == "none":
+            return "none"
+        else:
+            return {
+                "type": "function",
+                "function": {
+                    "name": tool_choice,
+                },
+            }
+
+    @classmethod
+    def convert_response_format(
+        cls, final_output_schema: AgentOutputSchemaBase | None
+    ) -> ResponseFormat | NotGiven:
+        if not final_output_schema or final_output_schema.is_plain_text():
+            return NOT_GIVEN
+
+        return {
+            "type": "json_schema",
+            "json_schema": {
+                "name": "final_output",
+                "strict": final_output_schema.is_strict_json_schema(),
+                "schema": final_output_schema.json_schema(),
+            },
+        }
+
+    @classmethod
+    def message_to_output_items(cls, message: ChatCompletionMessage) -> list[TResponseOutputItem]:
+        items: list[TResponseOutputItem] = []
+
+        message_item = ResponseOutputMessage(
+            id=FAKE_RESPONSES_ID,
+            content=[],
+            role="assistant",
+            type="message",
+            status="completed",
+        )
+        if message.content:
+            message_item.content.append(
+                ResponseOutputText(text=message.content, type="output_text", annotations=[])
+            )
+        if message.refusal:
+            message_item.content.append(
+                ResponseOutputRefusal(refusal=message.refusal, type="refusal")
+            )
+        if message.audio:
+            raise AgentsException("Audio is not currently supported")
+
+        if message_item.content:
+            items.append(message_item)
+
+        if message.tool_calls:
+            for tool_call in message.tool_calls:
+                items.append(
+                    ResponseFunctionToolCall(
+                        id=FAKE_RESPONSES_ID,
+                        call_id=tool_call.id,
+                        arguments=tool_call.function.arguments,
+                        name=tool_call.function.name,
+                        type="function_call",
+                    )
+                )
+
+        return items
+
+    @classmethod
+    def maybe_easy_input_message(cls, item: Any) -> EasyInputMessageParam | None:
+        if not isinstance(item, dict):
+            return None
+
+        keys = item.keys()
+        # EasyInputMessageParam only has these two keys
+        if keys != {"content", "role"}:
+            return None
+
+        role = item.get("role", None)
+        if role not in ("user", "assistant", "system", "developer"):
+            return None
+
+        if "content" not in item:
+            return None
+
+        return cast(EasyInputMessageParam, item)
+
+    @classmethod
+    def maybe_input_message(cls, item: Any) -> Message | None:
+        if (
+            isinstance(item, dict)
+            and item.get("type") == "message"
+            and item.get("role")
+            in (
+                "user",
+                "system",
+                "developer",
+            )
+        ):
+            return cast(Message, item)
+
+        return None
+
+    @classmethod
+    def maybe_file_search_call(cls, item: Any) -> ResponseFileSearchToolCallParam | None:
+        if isinstance(item, dict) and item.get("type") == "file_search_call":
+            return cast(ResponseFileSearchToolCallParam, item)
+        return None
+
+    @classmethod
+    def maybe_function_tool_call(cls, item: Any) -> ResponseFunctionToolCallParam | None:
+        if isinstance(item, dict) and item.get("type") == "function_call":
+            return cast(ResponseFunctionToolCallParam, item)
+        return None
+
+    @classmethod
+    def maybe_function_tool_call_output(
+        cls,
+        item: Any,
+    ) -> FunctionCallOutput | None:
+        if isinstance(item, dict) and item.get("type") == "function_call_output":
+            return cast(FunctionCallOutput, item)
+        return None
+
+    @classmethod
+    def maybe_item_reference(cls, item: Any) -> ItemReference | None:
+        if isinstance(item, dict) and item.get("type") == "item_reference":
+            return cast(ItemReference, item)
+        return None
+
+    @classmethod
+    def maybe_response_output_message(cls, item: Any) -> ResponseOutputMessageParam | None:
+        # ResponseOutputMessage is only used for messages with role assistant
+        if (
+            isinstance(item, dict)
+            and item.get("type") == "message"
+            and item.get("role") == "assistant"
+        ):
+            return cast(ResponseOutputMessageParam, item)
+        return None
+
+    @classmethod
+    def extract_text_content(
+        cls, content: str | Iterable[ResponseInputContentParam]
+    ) -> str | list[ChatCompletionContentPartTextParam]:
+        all_content = cls.extract_all_content(content)
+        if isinstance(all_content, str):
+            return all_content
+        out: list[ChatCompletionContentPartTextParam] = []
+        for c in all_content:
+            if c.get("type") == "text":
+                out.append(cast(ChatCompletionContentPartTextParam, c))
+        return out
+
+    @classmethod
+    def extract_all_content(
+        cls, content: str | Iterable[ResponseInputContentParam]
+    ) -> str | list[ChatCompletionContentPartParam]:
+        if isinstance(content, str):
+            return content
+        out: list[ChatCompletionContentPartParam] = []
+
+        for c in content:
+            if isinstance(c, dict) and c.get("type") == "input_text":
+                casted_text_param = cast(ResponseInputTextParam, c)
+                out.append(
+                    ChatCompletionContentPartTextParam(
+                        type="text",
+                        text=casted_text_param["text"],
+                    )
+                )
+            elif isinstance(c, dict) and c.get("type") == "input_image":
+                casted_image_param = cast(ResponseInputImageParam, c)
+                if "image_url" not in casted_image_param or not casted_image_param["image_url"]:
+                    raise UserError(
+                        f"Only image URLs are supported for input_image {casted_image_param}"
+                    )
+                out.append(
+                    ChatCompletionContentPartImageParam(
+                        type="image_url",
+                        image_url={
+                            "url": casted_image_param["image_url"],
+                            "detail": casted_image_param["detail"],
+                        },
+                    )
+                )
+            elif isinstance(c, dict) and c.get("type") == "input_file":
+                raise UserError(f"File uploads are not supported for chat completions {c}")
+            else:
+                raise UserError(f"Unknown content: {c}")
+        return out
+
+    @classmethod
+    def items_to_messages(
+        cls,
+        items: str | Iterable[TResponseInputItem],
+    ) -> list[ChatCompletionMessageParam]:
+        """
+        Convert a sequence of 'Item' objects into a list of ChatCompletionMessageParam.
+
+        Rules:
+        - EasyInputMessage or InputMessage (role=user) => ChatCompletionUserMessageParam
+        - EasyInputMessage or InputMessage (role=system) => ChatCompletionSystemMessageParam
+        - EasyInputMessage or InputMessage (role=developer) => ChatCompletionDeveloperMessageParam
+        - InputMessage (role=assistant) => Start or flush a ChatCompletionAssistantMessageParam
+        - response_output_message => Also produces/flushes a ChatCompletionAssistantMessageParam
+        - tool calls get attached to the *current* assistant message, or create one if none.
+        - tool outputs => ChatCompletionToolMessageParam
+        """
+
+        if isinstance(items, str):
+            return [
+                ChatCompletionUserMessageParam(
+                    role="user",
+                    content=items,
+                )
+            ]
+
+        result: list[ChatCompletionMessageParam] = []
+        current_assistant_msg: ChatCompletionAssistantMessageParam | None = None
+
+        def flush_assistant_message() -> None:
+            nonlocal current_assistant_msg
+            if current_assistant_msg is not None:
+                # The API doesn't support empty arrays for tool_calls
+                if not current_assistant_msg.get("tool_calls"):
+                    del current_assistant_msg["tool_calls"]
+                result.append(current_assistant_msg)
+                current_assistant_msg = None
+
+        def ensure_assistant_message() -> ChatCompletionAssistantMessageParam:
+            nonlocal current_assistant_msg
+            if current_assistant_msg is None:
+                current_assistant_msg = ChatCompletionAssistantMessageParam(role="assistant")
+                current_assistant_msg["tool_calls"] = []
+            return current_assistant_msg
+
+        for item in items:
+            # 1) Check easy input message
+            if easy_msg := cls.maybe_easy_input_message(item):
+                role = easy_msg["role"]
+                content = easy_msg["content"]
+
+                if role == "user":
+                    flush_assistant_message()
+                    msg_user: ChatCompletionUserMessageParam = {
+                        "role": "user",
+                        "content": cls.extract_all_content(content),
+                    }
+                    result.append(msg_user)
+                elif role == "system":
+                    flush_assistant_message()
+                    msg_system: ChatCompletionSystemMessageParam = {
+                        "role": "system",
+                        "content": cls.extract_text_content(content),
+                    }
+                    result.append(msg_system)
+                elif role == "developer":
+                    flush_assistant_message()
+                    msg_developer: ChatCompletionDeveloperMessageParam = {
+                        "role": "developer",
+                        "content": cls.extract_text_content(content),
+                    }
+                    result.append(msg_developer)
+                elif role == "assistant":
+                    flush_assistant_message()
+                    msg_assistant: ChatCompletionAssistantMessageParam = {
+                        "role": "assistant",
+                        "content": cls.extract_text_content(content),
+                    }
+                    result.append(msg_assistant)
+                else:
+                    raise UserError(f"Unexpected role in easy_input_message: {role}")
+
+            # 2) Check input message
+            elif in_msg := cls.maybe_input_message(item):
+                role = in_msg["role"]
+                content = in_msg["content"]
+                flush_assistant_message()
+
+                if role == "user":
+                    msg_user = {
+                        "role": "user",
+                        "content": cls.extract_all_content(content),
+                    }
+                    result.append(msg_user)
+                elif role == "system":
+                    msg_system = {
+                        "role": "system",
+                        "content": cls.extract_text_content(content),
+                    }
+                    result.append(msg_system)
+                elif role == "developer":
+                    msg_developer = {
+                        "role": "developer",
+                        "content": cls.extract_text_content(content),
+                    }
+                    result.append(msg_developer)
+                else:
+                    raise UserError(f"Unexpected role in input_message: {role}")
+
+            # 3) response output message => assistant
+            elif resp_msg := cls.maybe_response_output_message(item):
+                flush_assistant_message()
+                new_asst = ChatCompletionAssistantMessageParam(role="assistant")
+                contents = resp_msg["content"]
+
+                text_segments = []
+                for c in contents:
+                    if c["type"] == "output_text":
+                        text_segments.append(c["text"])
+                    elif c["type"] == "refusal":
+                        new_asst["refusal"] = c["refusal"]
+                    elif c["type"] == "output_audio":
+                        # Can't handle this, b/c chat completions expects an ID which we dont have
+                        raise UserError(
+                            f"Only audio IDs are supported for chat completions, but got: {c}"
+                        )
+                    else:
+                        raise UserError(f"Unknown content type in ResponseOutputMessage: {c}")
+
+                if text_segments:
+                    combined = "\n".join(text_segments)
+                    new_asst["content"] = combined
+
+                new_asst["tool_calls"] = []
+                current_assistant_msg = new_asst
+
+            # 4) function/file-search calls => attach to assistant
+            elif file_search := cls.maybe_file_search_call(item):
+                asst = ensure_assistant_message()
+                tool_calls = list(asst.get("tool_calls", []))
+                new_tool_call = ChatCompletionMessageToolCallParam(
+                    id=file_search["id"],
+                    type="function",
+                    function={
+                        "name": "file_search_call",
+                        "arguments": json.dumps(
+                            {
+                                "queries": file_search.get("queries", []),
+                                "status": file_search.get("status"),
+                            }
+                        ),
+                    },
+                )
+                tool_calls.append(new_tool_call)
+                asst["tool_calls"] = tool_calls
+
+            elif func_call := cls.maybe_function_tool_call(item):
+                asst = ensure_assistant_message()
+                tool_calls = list(asst.get("tool_calls", []))
+                arguments = func_call["arguments"] if func_call["arguments"] else "{}"
+                new_tool_call = ChatCompletionMessageToolCallParam(
+                    id=func_call["call_id"],
+                    type="function",
+                    function={
+                        "name": func_call["name"],
+                        "arguments": arguments,
+                    },
+                )
+                tool_calls.append(new_tool_call)
+                asst["tool_calls"] = tool_calls
+            # 5) function call output => tool message
+            elif func_output := cls.maybe_function_tool_call_output(item):
+                flush_assistant_message()
+                msg: ChatCompletionToolMessageParam = {
+                    "role": "tool",
+                    "tool_call_id": func_output["call_id"],
+                    "content": func_output["output"],
+                }
+                result.append(msg)
+
+            # 6) item reference => handle or raise
+            elif item_ref := cls.maybe_item_reference(item):
+                raise UserError(
+                    f"Encountered an item_reference, which is not supported: {item_ref}"
+                )
+
+            # 7) If we haven't recognized it => fail or ignore
+            else:
+                raise UserError(f"Unhandled item type or structure: {item}")
+
+        flush_assistant_message()
+        return result
+
+    @classmethod
+    def tool_to_openai(cls, tool: Tool) -> ChatCompletionToolParam:
+        if isinstance(tool, FunctionTool):
+            return {
+                "type": "function",
+                "function": {
+                    "name": tool.name,
+                    "description": tool.description or "",
+                    "parameters": tool.params_json_schema,
+                },
+            }
+
+        raise UserError(
+            f"Hosted tools are not supported with the ChatCompletions API. Got tool type: "
+            f"{type(tool)}, tool: {tool}"
+        )
+
+    @classmethod
+    def convert_handoff_tool(cls, handoff: Handoff[Any]) -> ChatCompletionToolParam:
+        return {
+            "type": "function",
+            "function": {
+                "name": handoff.tool_name,
+                "description": handoff.tool_description,
+                "parameters": handoff.input_json_schema,
+            },
+        }

agents/models/chatcmpl_helpers.py

@@ -0,0 +1,37 @@
+from __future__ import annotations
+
+from openai import AsyncOpenAI
+
+from ..model_settings import ModelSettings
+from ..version import __version__
+
+_USER_AGENT = f"Agents/Python {__version__}"
+HEADERS = {"User-Agent": _USER_AGENT}
+
+
+class ChatCmplHelpers:
+    @classmethod
+    def is_openai(cls, client: AsyncOpenAI):
+        return str(client.base_url).startswith("https://api.openai.com")
+
+    @classmethod
+    def get_store_param(cls, client: AsyncOpenAI, model_settings: ModelSettings) -> bool | None:
+        # Match the behavior of Responses where store is True when not given
+        default_store = True if cls.is_openai(client) else None
+        return model_settings.store if model_settings.store is not None else default_store
+
+    @classmethod
+    def get_stream_options_param(
+        cls, client: AsyncOpenAI, model_settings: ModelSettings, stream: bool
+    ) -> dict[str, bool] | None:
+        if not stream:
+            return None
+
+        default_include_usage = True if cls.is_openai(client) else None
+        include_usage = (
+            model_settings.include_usage
+            if model_settings.include_usage is not None
+            else default_include_usage
+        )
+        stream_options = {"include_usage": include_usage} if include_usage is not None else None
+        return stream_options