Agent Executor

AgentExecutor Class#

The AgentExecutor class is a core component of the dowse library. It utilizes emp_agents to build LLM powered processing steps in an executionpipeline. It is designed to facilitate the execution of commands by processing inputs through a series of steps, including prompt generation, tool invocation, and result aggregation.

Importing AgentExecutor#

To use the AgentExecutor class, you need to import it from the dowse library. Here is an example of how to import the AgentExecutor class:

from dowse import AgentExecutor

Initializing AgentExecutor#

The AgentExecutor class is initialized with a prompt, tools, and an executor. Here is an example of how to initialize the AgentExecutor class:

agent_executor = AgentExecutor(
    prompt="You are a helpful assistant.",
    tools=[],
    executor=Executor(),
)

Relative Imports#

The AgentExecutor can import its prompt, tools, and executor from the relative files to the initialiation path of the executor.

If you structure your project like this:

dowse/
    __init__.py
    agent_executor.py
    PROMPT.txt
    tools.py
    examples.py

Executing Commands#

The AgentExecutor class is executed by calling the execute method. Here is an example of how to execute a command:

from pydantic import BaseModel

from dowse import AgentExecutor
from dowse.models import AgentMessage

class InputType(BaseModel):
    name: str

class OutputType(BaseModel):
    age: int

executor = AgentExecutor[InputType, OutputType](
    prompt="Guess the person's age given their name."
)

output: AgentMessage[OutputType] = await executor.execute(InputType(name="John"))