letta-server/letta/schemas/agent.py

from datetime import datetime
from enum import Enum
from typing import Dict, List, Optional

from pydantic import BaseModel, Field, field_validator

from letta.schemas.block import CreateBlock
from letta.schemas.embedding_config import EmbeddingConfig
from letta.schemas.letta_base import LettaBase
from letta.schemas.llm_config import LLMConfig
from letta.schemas.memory import Memory
from letta.schemas.message import Message
from letta.schemas.openai.chat_completion_response import UsageStatistics
from letta.schemas.source import Source
from letta.schemas.tool import Tool
from letta.schemas.tool_rule import ToolRule


class BaseAgent(LettaBase, validate_assignment=True):
    __id_prefix__ = "agent"
    description: Optional[str] = Field(None, description="The description of the agent.")

    # metadata
    metadata_: Optional[Dict] = Field(None, description="The metadata of the agent.", alias="metadata_")
    user_id: Optional[str] = Field(None, description="The user id of the agent.")


class AgentType(str, Enum):
    """
    Enum to represent the type of agent.
    """

    memgpt_agent = "memgpt_agent"
    split_thread_agent = "split_thread_agent"
    o1_agent = "o1_agent"


class PersistedAgentState(BaseAgent, validate_assignment=True):
    # NOTE: this has been changed to represent the data stored in the ORM, NOT what is passed around internally or returned to the user
    id: str = BaseAgent.generate_id_field()
    name: str = Field(..., description="The name of the agent.")
    created_at: datetime = Field(..., description="The datetime the agent was created.", default_factory=datetime.now)

    # in-context memory
    message_ids: Optional[List[str]] = Field(default=None, description="The ids of the messages in the agent's in-context memory.")

    # tools
    # TODO: move to ORM mapping
    tool_names: List[str] = Field(..., description="The tools used by the agent.")

    # tool rules
    tool_rules: Optional[List[ToolRule]] = Field(default=None, description="The list of tool rules.")

    # system prompt
    system: str = Field(..., description="The system prompt used by the agent.")

    # agent configuration
    agent_type: AgentType = Field(..., description="The type of agent.")

    # llm information
    llm_config: LLMConfig = Field(..., description="The LLM configuration used by the agent.")
    embedding_config: EmbeddingConfig = Field(..., description="The embedding configuration used by the agent.")

    class Config:
        arbitrary_types_allowed = True
        validate_assignment = True


class AgentState(PersistedAgentState):
    """
    Representation of an agent's state. This is the state of the agent at a given time, and is persisted in the DB backend. The state has all the information needed to recreate a persisted agent.

    Parameters:
        id (str): The unique identifier of the agent.
        name (str): The name of the agent (must be unique to the user).
        created_at (datetime): The datetime the agent was created.
        message_ids (List[str]): The ids of the messages in the agent's in-context memory.
        memory (Memory): The in-context memory of the agent.
        tools (List[str]): The tools used by the agent. This includes any memory editing functions specified in `memory`.
        system (str): The system prompt used by the agent.
        llm_config (LLMConfig): The LLM configuration used by the agent.
        embedding_config (EmbeddingConfig): The embedding configuration used by the agent.

    """

    # NOTE: this is what is returned to the client and also what is used to initialize `Agent`

    # This is an object representing the in-process state of a running `Agent`
    # Field in this object can be theoretically edited by tools, and will be persisted by the ORM
    memory: Memory = Field(..., description="The in-context memory of the agent.")
    tools: List[Tool] = Field(..., description="The tools used by the agent.")
    sources: List[Source] = Field(..., description="The sources used by the agent.")
    tags: List[str] = Field(..., description="The tags associated with the agent.")
    # TODO: add in context message objects

    def to_persisted_agent_state(self) -> PersistedAgentState:
        # turn back into persisted agent
        data = self.model_dump()
        del data["memory"]
        del data["tools"]
        del data["sources"]
        del data["tags"]
        return PersistedAgentState(**data)


class CreateAgent(BaseAgent):  #
    # all optional as server can generate defaults
    name: Optional[str] = Field(None, description="The name of the agent.")
    message_ids: Optional[List[str]] = Field(None, description="The ids of the messages in the agent's in-context memory.")

    # memory creation
    memory_blocks: List[CreateBlock] = Field(
        # [CreateHuman(), CreatePersona()], description="The blocks to create in the agent's in-context memory."
        ...,
        description="The blocks to create in the agent's in-context memory.",
    )

    tools: Optional[List[str]] = Field(None, description="The tools used by the agent.")
    tool_rules: Optional[List[ToolRule]] = Field(None, description="The tool rules governing the agent.")
    tags: Optional[List[str]] = Field(None, description="The tags associated with the agent.")
    system: Optional[str] = Field(None, description="The system prompt used by the agent.")
    agent_type: Optional[AgentType] = Field(None, description="The type of agent.")
    llm_config: Optional[LLMConfig] = Field(None, description="The LLM configuration used by the agent.")
    embedding_config: Optional[EmbeddingConfig] = Field(None, description="The embedding configuration used by the agent.")
    # Note: if this is None, then we'll populate with the standard "more human than human" initial message sequence
    # If the client wants to make this empty, then the client can set the arg to an empty list
    initial_message_sequence: Optional[List[Message]] = Field(
        None, description="The initial set of messages to put in the agent's in-context memory."
    )

    @field_validator("name")
    @classmethod
    def validate_name(cls, name: str) -> str:
        """Validate the requested new agent name (prevent bad inputs)"""

        import re

        if not name:
            # don't check if not provided
            return name

        # TODO: this check should also be added to other model (e.g. User.name)
        # Length check
        if not (1 <= len(name) <= 50):
            raise ValueError("Name length must be between 1 and 50 characters.")

        # Regex for allowed characters (alphanumeric, spaces, hyphens, underscores)
        if not re.match("^[A-Za-z0-9 _-]+$", name):
            raise ValueError("Name contains invalid characters.")

        # Further checks can be added here...
        # TODO

        return name


class UpdateAgentState(BaseAgent):
    id: str = Field(..., description="The id of the agent.")
    name: Optional[str] = Field(None, description="The name of the agent.")
    tool_names: Optional[List[str]] = Field(None, description="The tools used by the agent.")
    tags: Optional[List[str]] = Field(None, description="The tags associated with the agent.")
    system: Optional[str] = Field(None, description="The system prompt used by the agent.")
    llm_config: Optional[LLMConfig] = Field(None, description="The LLM configuration used by the agent.")
    embedding_config: Optional[EmbeddingConfig] = Field(None, description="The embedding configuration used by the agent.")

    # TODO: determine if these should be editable via this schema?
    message_ids: Optional[List[str]] = Field(None, description="The ids of the messages in the agent's in-context memory.")


class AgentStepResponse(BaseModel):
    messages: List[Message] = Field(..., description="The messages generated during the agent's step.")
    heartbeat_request: bool = Field(..., description="Whether the agent requested a heartbeat (i.e. follow-up execution).")
    function_failed: bool = Field(..., description="Whether the agent step ended because a function call failed.")
    in_context_memory_warning: bool = Field(
        ..., description="Whether the agent step ended because the in-context memory is near its limit."
    )
    usage: UsageStatistics = Field(..., description="Usage statistics of the LLM call during the agent's step.")