// Format A: string (works)
{"role": "tool", "tool_call_id": "call_123", "content": "22.5°C, partly cloudy"}

// Format B: list of content parts (broken — content silently dropped)
{"role": "tool", "tool_call_id": "call_123", "content": [{"type": "text", "text": "22.5°C, partly cloudy"}]}

---

curl -s http://localhost:8090/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $LITELLM_KEY" \
  -d '{
    "model": "zai/glm-5.1",
    "messages": [
      {"role": "user", "content": [{"type": "text", "text": "What is the temperature in Tokyo?"}]},
      {"role": "assistant", "content": [{"type": "text", "text": "Let me check."}],
       "tool_calls": [{"id": "call_1", "type": "function", "function": {"name": "get_temp", "arguments": "{\"city\":\"Tokyo\"}"}}]},
      {"role": "tool", "tool_call_id": "call_1",
       "content": [{"type": "text", "text": "The temperature in Tokyo is 22.5°C, partly cloudy."}]}
    ],
    "tools": [{"type": "function", "function": {"name": "get_temp", "description": "Get temperature", "parameters": {"type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"]}}}],
    "temperature": 0.1
  }'

---

# litellm/llms/zai/chat/transformation.py

from typing import Any, Coroutine, List, Literal, Optional, Tuple, Union, overload

from litellm.secret_managers.main import get_secret_str
from litellm.types.llms.openai import AllMessageValues, ChatCompletionToolParam

from ...openai.chat.gpt_transformation import OpenAIGPTConfig

ZAI_API_BASE = "https://api.z.ai/api/paas/v4"


def _flatten_content_parts(content):
    """
    Flatten OpenAI multi-part content to a plain string.

    The OpenAI API allows tool message content as either a string or a list
    of content parts like [{"type": "text", "text": "..."}].  GLM's chat
    template checks ``m.content is string`` and silently drops list-format
    content (see vllm-project/vllm#39614).  This helper normalises both
    forms to a plain string so the content always reaches the model.
    """
    if isinstance(content, str) or content is None:
        return content
    if isinstance(content, list):
        parts = []
        for part in content:
            if isinstance(part, dict):
                text = part.get("text")
                if text:
                    parts.append(text)
            elif isinstance(part, str):
                parts.append(part)
        return "\n".join(parts) if parts else ""
    return content


class ZAIChatConfig(OpenAIGPTConfig):
    # ... existing methods unchanged ...

    # fmt: off
    @overload
    def _transform_messages(
        self, messages: List[AllMessageValues], model: str, is_async: Literal[True]
    ) -> Coroutine[Any, Any, List[AllMessageValues]]: ...

    @overload
    def _transform_messages(
        self, messages: List[AllMessageValues], model: str, is_async: Literal[False] = False,
    ) -> List[AllMessageValues]: ...
    # fmt: on

    def _transform_messages(
        self, messages: List[AllMessageValues], model: str, is_async: bool = False
    ) -> Union[List[AllMessageValues], Coroutine[Any, Any, List[AllMessageValues]]]:
        # Flatten list-format content in tool and assistant messages
        for message in messages:
            role = message.get("role")
            content = message.get("content")
            if role == "tool" and isinstance(content, list):
                message["content"] = _flatten_content_parts(content)
            elif role == "assistant" and isinstance(content, list):
                message["content"] = _flatten_content_parts(content)

        # Delegate to parent for user-message image-URL transforms, etc.
        return super()._transform_messages(messages, model, is_async=is_async)

    # ... rest of class unchanged ...