MemGPT/letta/schemas/letta_message.py

371 lines
15 KiB
Python

import json
from datetime import datetime, timezone
from enum import Enum
from typing import Annotated, List, Literal, Optional, Union
from pydantic import BaseModel, Field, field_serializer, field_validator
from letta.schemas.letta_message_content import (
LettaAssistantMessageContentUnion,
LettaUserMessageContentUnion,
get_letta_assistant_message_content_union_str_json_schema,
get_letta_user_message_content_union_str_json_schema,
)
# ---------------------------
# Letta API Messaging Schemas
# ---------------------------
class MessageType(str, Enum):
system_message = "system_message"
user_message = "user_message"
assistant_message = "assistant_message"
reasoning_message = "reasoning_message"
hidden_reasoning_message = "hidden_reasoning_message"
tool_call_message = "tool_call_message"
tool_return_message = "tool_return_message"
class LettaMessage(BaseModel):
"""
Base class for simplified Letta message response type. This is intended to be used for developers
who want the internal monologue, tool calls, and tool returns in a simplified format that does not
include additional information other than the content and timestamp.
Args:
id (str): The ID of the message
date (datetime): The date the message was created in ISO format
name (Optional[str]): The name of the sender of the message
message_type (MessageType): The type of the message
otid (Optional[str]): The offline threading id associated with this message
sender_id (Optional[str]): The id of the sender of the message, can be an identity id or agent id
"""
id: str
date: datetime
name: Optional[str] = None
message_type: MessageType = Field(..., description="The type of the message.")
otid: Optional[str] = None
sender_id: Optional[str] = None
@field_serializer("date")
def serialize_datetime(self, dt: datetime, _info):
"""
Remove microseconds since it seems like we're inconsistent with getting them
TODO: figure out why we don't always get microseconds (get_utc_time() does)
"""
if dt.tzinfo is None or dt.tzinfo.utcoffset(dt) is None:
dt = dt.replace(tzinfo=timezone.utc)
return dt.isoformat(timespec="seconds")
class SystemMessage(LettaMessage):
"""
A message generated by the system. Never streamed back on a response, only used for cursor pagination.
Args:
id (str): The ID of the message
date (datetime): The date the message was created in ISO format
name (Optional[str]): The name of the sender of the message
content (str): The message content sent by the system
"""
message_type: Literal[MessageType.system_message] = Field(MessageType.system_message, description="The type of the message.")
content: str = Field(..., description="The message content sent by the system")
class UserMessage(LettaMessage):
"""
A message sent by the user. Never streamed back on a response, only used for cursor pagination.
Args:
id (str): The ID of the message
date (datetime): The date the message was created in ISO format
name (Optional[str]): The name of the sender of the message
content (Union[str, List[LettaUserMessageContentUnion]]): The message content sent by the user (can be a string or an array of multi-modal content parts)
"""
message_type: Literal[MessageType.user_message] = Field(MessageType.user_message, description="The type of the message.")
content: Union[str, List[LettaUserMessageContentUnion]] = Field(
...,
description="The message content sent by the user (can be a string or an array of multi-modal content parts)",
json_schema_extra=get_letta_user_message_content_union_str_json_schema(),
)
class ReasoningMessage(LettaMessage):
"""
Representation of an agent's internal reasoning.
Args:
id (str): The ID of the message
date (datetime): The date the message was created in ISO format
name (Optional[str]): The name of the sender of the message
source (Literal["reasoner_model", "non_reasoner_model"]): Whether the reasoning
content was generated natively by a reasoner model or derived via prompting
reasoning (str): The internal reasoning of the agent
signature (Optional[str]): The model-generated signature of the reasoning step
"""
message_type: Literal[MessageType.reasoning_message] = Field(MessageType.reasoning_message, description="The type of the message.")
source: Literal["reasoner_model", "non_reasoner_model"] = "non_reasoner_model"
reasoning: str
signature: Optional[str] = None
class HiddenReasoningMessage(LettaMessage):
"""
Representation of an agent's internal reasoning where reasoning content
has been hidden from the response.
Args:
id (str): The ID of the message
date (datetime): The date the message was created in ISO format
name (Optional[str]): The name of the sender of the message
state (Literal["redacted", "omitted"]): Whether the reasoning
content was redacted by the provider or simply omitted by the API
hidden_reasoning (Optional[str]): The internal reasoning of the agent
"""
message_type: Literal[MessageType.hidden_reasoning_message] = Field(
MessageType.hidden_reasoning_message, description="The type of the message."
)
state: Literal["redacted", "omitted"]
hidden_reasoning: Optional[str] = None
class ToolCall(BaseModel):
name: str
arguments: str
tool_call_id: str
class ToolCallDelta(BaseModel):
name: Optional[str] = None
arguments: Optional[str] = None
tool_call_id: Optional[str] = None
def model_dump(self, *args, **kwargs):
"""
This is a workaround to exclude None values from the JSON dump since the
OpenAI style of returning chunks doesn't include keys with null values.
"""
kwargs["exclude_none"] = True
return super().model_dump(*args, **kwargs)
def json(self, *args, **kwargs):
return json.dumps(self.model_dump(exclude_none=True), *args, **kwargs)
class ToolCallMessage(LettaMessage):
"""
A message representing a request to call a tool (generated by the LLM to trigger tool execution).
Args:
id (str): The ID of the message
date (datetime): The date the message was created in ISO format
name (Optional[str]): The name of the sender of the message
tool_call (Union[ToolCall, ToolCallDelta]): The tool call
"""
message_type: Literal[MessageType.tool_call_message] = Field(MessageType.tool_call_message, description="The type of the message.")
tool_call: Union[ToolCall, ToolCallDelta]
def model_dump(self, *args, **kwargs):
"""
Handling for the ToolCallDelta exclude_none to work correctly
"""
kwargs["exclude_none"] = True
data = super().model_dump(*args, **kwargs)
if isinstance(data["tool_call"], dict):
data["tool_call"] = {k: v for k, v in data["tool_call"].items() if v is not None}
return data
class Config:
json_encoders = {
ToolCallDelta: lambda v: v.model_dump(exclude_none=True),
ToolCall: lambda v: v.model_dump(exclude_none=True),
}
@field_validator("tool_call", mode="before")
@classmethod
def validate_tool_call(cls, v):
"""
Casts dicts into ToolCallMessage objects. Without this extra validator, Pydantic will throw
an error if 'name' or 'arguments' are None instead of properly casting to ToolCallDelta
instead of ToolCall.
"""
if isinstance(v, dict):
if "name" in v and "arguments" in v and "tool_call_id" in v:
return ToolCall(name=v["name"], arguments=v["arguments"], tool_call_id=v["tool_call_id"])
elif "name" in v or "arguments" in v or "tool_call_id" in v:
return ToolCallDelta(name=v.get("name"), arguments=v.get("arguments"), tool_call_id=v.get("tool_call_id"))
else:
raise ValueError("tool_call must contain either 'name' or 'arguments'")
return v
class ToolReturnMessage(LettaMessage):
"""
A message representing the return value of a tool call (generated by Letta executing the requested tool).
Args:
id (str): The ID of the message
date (datetime): The date the message was created in ISO format
name (Optional[str]): The name of the sender of the message
tool_return (str): The return value of the tool
status (Literal["success", "error"]): The status of the tool call
tool_call_id (str): A unique identifier for the tool call that generated this message
stdout (Optional[List(str)]): Captured stdout (e.g. prints, logs) from the tool invocation
stderr (Optional[List(str)]): Captured stderr from the tool invocation
"""
message_type: Literal[MessageType.tool_return_message] = Field(MessageType.tool_return_message, description="The type of the message.")
tool_return: str
status: Literal["success", "error"]
tool_call_id: str
stdout: Optional[List[str]] = None
stderr: Optional[List[str]] = None
class AssistantMessage(LettaMessage):
"""
A message sent by the LLM in response to user input. Used in the LLM context.
Args:
id (str): The ID of the message
date (datetime): The date the message was created in ISO format
name (Optional[str]): The name of the sender of the message
content (Union[str, List[LettaAssistantMessageContentUnion]]): The message content sent by the agent (can be a string or an array of content parts)
"""
message_type: Literal[MessageType.assistant_message] = Field(MessageType.assistant_message, description="The type of the message.")
content: Union[str, List[LettaAssistantMessageContentUnion]] = Field(
...,
description="The message content sent by the agent (can be a string or an array of content parts)",
json_schema_extra=get_letta_assistant_message_content_union_str_json_schema(),
)
# NOTE: use Pydantic's discriminated unions feature: https://docs.pydantic.dev/latest/concepts/unions/#discriminated-unions
LettaMessageUnion = Annotated[
Union[SystemMessage, UserMessage, ReasoningMessage, HiddenReasoningMessage, ToolCallMessage, ToolReturnMessage, AssistantMessage],
Field(discriminator="message_type"),
]
def create_letta_message_union_schema():
return {
"oneOf": [
{"$ref": "#/components/schemas/SystemMessage"},
{"$ref": "#/components/schemas/UserMessage"},
{"$ref": "#/components/schemas/ReasoningMessage"},
{"$ref": "#/components/schemas/HiddenReasoningMessage"},
{"$ref": "#/components/schemas/ToolCallMessage"},
{"$ref": "#/components/schemas/ToolReturnMessage"},
{"$ref": "#/components/schemas/AssistantMessage"},
],
"discriminator": {
"propertyName": "message_type",
"mapping": {
"system_message": "#/components/schemas/SystemMessage",
"user_message": "#/components/schemas/UserMessage",
"reasoning_message": "#/components/schemas/ReasoningMessage",
"hidden_reasoning_message": "#/components/schemas/HiddenReasoningMessage",
"tool_call_message": "#/components/schemas/ToolCallMessage",
"tool_return_message": "#/components/schemas/ToolReturnMessage",
"assistant_message": "#/components/schemas/AssistantMessage",
},
},
}
# --------------------------
# Message Update API Schemas
# --------------------------
class UpdateSystemMessage(BaseModel):
message_type: Literal["system_message"] = "system_message"
content: str = Field(
..., description="The message content sent by the system (can be a string or an array of multi-modal content parts)"
)
class UpdateUserMessage(BaseModel):
message_type: Literal["user_message"] = "user_message"
content: Union[str, List[LettaUserMessageContentUnion]] = Field(
...,
description="The message content sent by the user (can be a string or an array of multi-modal content parts)",
json_schema_extra=get_letta_user_message_content_union_str_json_schema(),
)
class UpdateReasoningMessage(BaseModel):
reasoning: str
message_type: Literal["reasoning_message"] = "reasoning_message"
class UpdateAssistantMessage(BaseModel):
message_type: Literal["assistant_message"] = "assistant_message"
content: Union[str, List[LettaAssistantMessageContentUnion]] = Field(
...,
description="The message content sent by the assistant (can be a string or an array of content parts)",
json_schema_extra=get_letta_assistant_message_content_union_str_json_schema(),
)
LettaMessageUpdateUnion = Annotated[
Union[UpdateSystemMessage, UpdateUserMessage, UpdateReasoningMessage, UpdateAssistantMessage],
Field(discriminator="message_type"),
]
# --------------------------
# Deprecated Message Schemas
# --------------------------
class LegacyFunctionCallMessage(LettaMessage):
function_call: str
class LegacyFunctionReturn(LettaMessage):
"""
A message representing the return value of a function call (generated by Letta executing the requested function).
Args:
function_return (str): The return value of the function
status (Literal["success", "error"]): The status of the function call
id (str): The ID of the message
date (datetime): The date the message was created in ISO format
function_call_id (str): A unique identifier for the function call that generated this message
stdout (Optional[List(str)]): Captured stdout (e.g. prints, logs) from the function invocation
stderr (Optional[List(str)]): Captured stderr from the function invocation
"""
message_type: Literal["function_return"] = "function_return"
function_return: str
status: Literal["success", "error"]
function_call_id: str
stdout: Optional[List[str]] = None
stderr: Optional[List[str]] = None
class LegacyInternalMonologue(LettaMessage):
"""
Representation of an agent's internal monologue.
Args:
internal_monologue (str): The internal monologue of the agent
id (str): The ID of the message
date (datetime): The date the message was created in ISO format
"""
message_type: Literal["internal_monologue"] = "internal_monologue"
internal_monologue: str
LegacyLettaMessage = Union[LegacyInternalMonologue, AssistantMessage, LegacyFunctionCallMessage, LegacyFunctionReturn]