基类只定了接口（libs/core/langchain_core/language_models/chat_models.py:2338）：

# libs/core/langchain_core/language_models/chat_models.py:2338
def bind_tools(
    self,
    tools: Sequence[dict[str, Any] | type | Callable[..., Any] | BaseTool],  # 四种形态都收
    *,
    tool_choice: str | None = None,
    **kwargs: Any,
) -> Runnable[LanguageModelInput, AIMessage]:
    """Bind tools to the model."""
    raise NotImplementedError        # 具体翻译格式，各厂商包自己实现

看真实现——OpenAI 版（libs/partners/openai/langchain_openai/chat_models/base.py:2157）：

# libs/partners/openai/langchain_openai/chat_models/base.py:2157（裁剪）
def bind_tools(self, tools, *, tool_choice=None, strict=None, **kwargs):
    formatted_tools = [
        convert_to_openai_tool(tool, strict=strict) for tool in tools   # :2203 ★统一翻译
    ]
    ...
    if tool_choice:                                  # "auto"/"any"/指定工具名/none
        ...
    return super().bind(tools=formatted_tools, **kwargs)   # :2263 ★挂参数，返回新 Runnable

模型决定用工具时，回复的 AIMessage（libs/core/langchain_core/messages/ai.py:160）里 tool_calls 字段非空：

# libs/core/langchain_core/messages/ai.py:160
class AIMessage(BaseMessage):
    """Message from an AI. ..."""

    tool_calls: list[ToolCall] = Field(default_factory=list)          # :170 ★工单列表
    invalid_tool_calls: list[InvalidToolCall] = Field(default_factory=list)  # :173 解析失败的工单
    usage_metadata: UsageMetadata | None = None                       # :176 token 用量
    type: Literal["ai"] = "ai"

每张工单是个 TypedDict——ToolCall（libs/core/langchain_core/messages/tool.py:206）：

# libs/core/langchain_core/messages/tool.py:206
class ToolCall(TypedDict):
    """Represents an AI's request to call a tool.

    Example:
        {"name": "foo", "args": {"a": 1}, "id": "123"}
    """
    name: str               # 调哪件工具（Day13 的品名）
    args: dict[str, Any]    # ★参数：已解析成标准 dict（不管厂商原始格式是啥）
    id: str | None          # ★工单号：回执对账全靠它
    type: NotRequired[Literal["tool_call"]]

你执行完工具，用第四种消息还给模型——ToolMessage（libs/core/langchain_core/messages/tool.py:26）：

# libs/core/langchain_core/messages/tool.py:26（裁剪）
class ToolMessage(BaseMessage, ToolOutputMixin):
    """Message for passing the result of executing a tool back to a model.

    Example:
        ToolMessage(content="42", tool_call_id="call_Jja7J89XsjrOLA5r...")
    """
    tool_call_id: str                                  # :67 ★必填！对应 ToolCall.id
    artifact: Any = None                               # :73 原始产物（不发给模型）
    status: Literal["success", "error"] = "success"   # :81 执行成败标记
    type: Literal["tool"] = "tool"

手工闭环要自己取 id、造 ToolMessage？不用。Day13 埋的伏笔兑现：把整个 ToolCall 直接丢给工具的 invoke。看 libs/core/langchain_core/tools/base.py:731 → _prep_run_args（base.py:1320）：

# libs/core/langchain_core/tools/base.py:1320（裁剪）
def _prep_run_args(value, config, **kwargs):
    if _is_tool_call(value):                              # ① 认出这是一张工单
        tool_call_id: str | None = value["id"]            # ② ★抽出工单号
        tool_input = value["args"].copy()                 # ③ 参数就是 args dict
    else:
        tool_call_id = None                               #    普通调用：没有工单号
        tool_input = value
    return (tool_input, dict(..., tool_call_id=tool_call_id, **kwargs))

执行完，出口处 _format_output（base.py:1358）决定返回什么：

# libs/core/langchain_core/tools/base.py:1358（裁剪）
def _format_output(content, artifact, tool_call_id, name, status):
    if isinstance(content, ToolOutputMixin) or tool_call_id is None:
        return content                                    # 普通调用 → 原样返回结果
    ...
    return ToolMessage(content,                           # ★带工单号调用 → 自动包成回执
                       artifact=artifact,
                       tool_call_id=tool_call_id,
                       name=name,
                       status=status)