A message is one structured entry in a conversation. Its role identifies the source or purpose, and its content normally holds text. Messages form an ordered history that a session stores and a model request can use as context. section 4.1 section 4.0
The core fields support both conversation and tool execution:
- role distinguishes user input, assistant responses, system instructions, and tool results. section 4.1
- content holds text, but can be null when an assistant requests a tool call instead of answering. section 4.1
- tool_calls records requested functions, their arguments, and call IDs. A tool result uses tool_call_id to link back to its request, including when multiple calls run in parallel. section 4.1
- timestamp records when the message was created; the service contract also includes an optional name field. section 4.1 section 4.2
Listing 4.1 Message dataclass representing a single entry in the conversation
@dataclass
class Message:
role: str
content: Optional[str] = None
tool_calls: Optional[List[ToolCall]] = None #A
tool_call_id: Optional[str] = None #A
timestamp: datetime = field(default_factory=datetime.utcnow)
@dataclass
class ToolCall:
id: str #B
type: Literal["function"]
function: Function
@dataclass
class Function:
name: str
arguments: str #C
The Python data model makes content and tool metadata optional and nests each function request inside a ToolCall. section 4.1
Appointment availability
A patient asks about openings. The assistant sends a message with null content and a check_availability tool call. The workflow executes it and returns a tool message with the matching call ID and available slots; the assistant then answers the patient. section 4.1
A workflow records and retrieves messages through the Session Service, then supplies conversation history in a Model Service chat request. Context-window limits mean a long history may need to be restricted before the next request. section 4.2 section 3.1.6 section 4.0
Listing 4.7 The Message structure and its nested types
message Message {
string role = 1;
optional string content = 2;
repeated ToolCall tool_calls = 3;
optional string tool_call_id = 4;
optional string name = 5;
int64 timestamp = 6;
}
message ToolCall {
string id = 1; #A
string type = 2;
Function function = 3;
}
message Function {
string name = 1;
string arguments = 2; #B
}
The Protocol Buffer representation carries message fields and nested tool-call types across the service boundary. section 4.2
Using an OpenAI-style canonical message format keeps stored history and model requests familiar and minimizes translation for OpenAI and OpenAI-compatible vLLM endpoints. Provider adapters translate between this representation and other provider formats. section 3.2.3
