Function run

methods/letta/cli/cli.py:75–329

Start chatting with an Letta agent Example usage: `letta run --agent myagent --data-source mydata --persona mypersona --human myhuman --model gpt-3.5-turbo` :param persona: Specify persona :param agent: Specify agent name (will load existing state if the agent exists, or create a new o

(
    persona: Annotated[Optional[str], typer.Option(help="Specify persona")] = None,
    agent: Annotated[Optional[str], typer.Option(help="Specify agent name")] = None,
    human: Annotated[Optional[str], typer.Option(help="Specify human")] = None,
    system: Annotated[Optional[str], typer.Option(help="Specify system prompt (raw text)")] = None,
    system_file: Annotated[Optional[str], typer.Option(help="Specify raw text file containing system prompt")] = None,
    # model flags
    model: Annotated[Optional[str], typer.Option(help="Specify the LLM model")] = None,
    model_wrapper: Annotated[Optional[str], typer.Option(help="Specify the LLM model wrapper")] = None,
    model_endpoint: Annotated[Optional[str], typer.Option(help="Specify the LLM model endpoint")] = None,
    model_endpoint_type: Annotated[Optional[str], typer.Option(help="Specify the LLM model endpoint type")] = None,
    context_window: Annotated[
        Optional[int], typer.Option(help="The context window of the LLM you are using (e.g. 8k for most Mistral 7B variants)")
    ] = None,
    core_memory_limit: Annotated[
        Optional[int], typer.Option(help="The character limit to each core-memory section (human/persona).")
    ] = CORE_MEMORY_BLOCK_CHAR_LIMIT,
    # other
    first: Annotated[bool, typer.Option(help="Use --first to send the first message in the sequence")] = False,
    strip_ui: Annotated[bool, typer.Option(help="Remove all the bells and whistles in CLI output (helpful for testing)")] = False,
    debug: Annotated[bool, typer.Option(help="Use --debug to enable debugging output")] = False,
    no_verify: Annotated[bool, typer.Option(help="Bypass message verification")] = False,
    yes: Annotated[bool, typer.Option("-y", help="Skip confirmation prompt and use defaults")] = False,
    # streaming
    stream: Annotated[bool, typer.Option(help="Enables message streaming in the CLI (if the backend supports it)")] = False,
    # whether or not to put the inner thoughts inside the function args
    no_content: Annotated[
        OptionState, typer.Option(help="Set to 'yes' for LLM APIs that omit the `content` field during tool calling")
    ] = OptionState.DEFAULT,
)

def run(
    persona: Annotated[Optional[str], typer.Option(help="Specify persona")] = None,
    agent: Annotated[Optional[str], typer.Option(help="Specify agent name")] = None,
    human: Annotated[Optional[str], typer.Option(help="Specify human")] = None,
    system: Annotated[Optional[str], typer.Option(help="Specify system prompt (raw text)")] = None,
    system_file: Annotated[Optional[str], typer.Option(help="Specify raw text file containing system prompt")] = None,
    # model flags
    model: Annotated[Optional[str], typer.Option(help="Specify the LLM model")] = None,
    model_wrapper: Annotated[Optional[str], typer.Option(help="Specify the LLM model wrapper")] = None,
    model_endpoint: Annotated[Optional[str], typer.Option(help="Specify the LLM model endpoint")] = None,
    model_endpoint_type: Annotated[Optional[str], typer.Option(help="Specify the LLM model endpoint type")] = None,
    context_window: Annotated[
        Optional[int], typer.Option(help="The context window of the LLM you are using (e.g. 8k for most Mistral 7B variants)")
    ] = None,
    core_memory_limit: Annotated[
        Optional[int], typer.Option(help="The character limit to each core-memory section (human/persona).")
    ] = CORE_MEMORY_BLOCK_CHAR_LIMIT,
    # other
    first: Annotated[bool, typer.Option(help="Use --first to send the first message in the sequence")] = False,
    strip_ui: Annotated[bool, typer.Option(help="Remove all the bells and whistles in CLI output (helpful for testing)")] = False,
    debug: Annotated[bool, typer.Option(help="Use --debug to enable debugging output")] = False,
    no_verify: Annotated[bool, typer.Option(help="Bypass message verification")] = False,
    yes: Annotated[bool, typer.Option("-y", help="Skip confirmation prompt and use defaults")] = False,
    # streaming
    stream: Annotated[bool, typer.Option(help="Enables message streaming in the CLI (if the backend supports it)")] = False,
    # whether or not to put the inner thoughts inside the function args
    no_content: Annotated[
        OptionState, typer.Option(help="Set to 'yes' for LLM APIs that omit the `content` field during tool calling")
    ] = OptionState.DEFAULT,
):
    """Start chatting with an Letta agent

    Example usage: `letta run --agent myagent --data-source mydata --persona mypersona --human myhuman --model gpt-3.5-turbo`

    :param persona: Specify persona
    :param agent: Specify agent name (will load existing state if the agent exists, or create a new one with that name)
    :param human: Specify human
    :param model: Specify the LLM model

    """

    # setup logger
    # TODO: remove Utils Debug after global logging is complete.
    utils.DEBUG = debug
    # TODO: add logging command line options for runtime log level

    from letta.server.server import logger as server_logger

    if debug:
        logger.setLevel(logging.DEBUG)
        server_logger.setLevel(logging.DEBUG)
    else:
        logger.setLevel(logging.CRITICAL)
        server_logger.setLevel(logging.CRITICAL)

    # load config file
    config = LettaConfig.load()