Class Message

ms_agent/llm/utils.py:28–96

@dataclass
class Message:
    role: Literal['system', 'user', 'assistant', 'tool']

    content: Union[str, List[Dict[str, str]]] = ''

    tool_calls: List[ToolCall] = field(default_factory=list)

    tool_call_id: Optional[str] = None

    # Also defined in OpenAI message
    name: Optional[str] = None

    # needed for output
    reasoning_content: str = ''

    # request id
    id: str = ''

    # continue generation mode
    partial: bool = False
    prefix: bool = False

    # UI resources from mcp result
    resources: List[Dict[str, str]] = field(default_factory=list)

    # usage
    completion_tokens: int = 0
    prompt_tokens: int = 0

    # tokens that hit existing cache (billed at reduced rate like 0.1x)
    cached_tokens: int = 0
    # tokens used to create new cache (explicit cache only, billed at higher rate like 1.25x)
    cache_creation_input_tokens: int = 0

    api_calls: int = 1

    # Knowledge search (sirchmunk) related fields
    # searching_detail: Search process logs and metadata for frontend display
    searching_detail: Dict[str, Any] = field(default_factory=dict)
    # search_result: Raw search results to be used as context for next LLM turn
    search_result: List[Dict[str, Any]] = field(default_factory=list)

    def to_dict(self):
        return asdict(self)

    def to_dict_clean(self):
        raw_dict = asdict(self)
        if raw_dict.get('tool_calls'):
            for idx, tool_call in enumerate(raw_dict['tool_calls']):
                try:
                    if tool_call['arguments']:
                        json.loads(tool_call['arguments'])
                except Exception:
                    tool_call['arguments'] = '{}'
                raw_dict['tool_calls'][idx] = {
                    'id': tool_call['id'],
                    'type': tool_call['type'],
                    'function': {