The AgentBuilder class provides a configurable interface for setting up your agent’s behavior. Below is the default configuration, optimized for OpenAI users (due to its popularity):
class AgentBuilder:
def __init__(self,
planner_provider="openai",
planner_model="o3-mini",
writer_provider="openai",
writer_model="gpt-4.1-nano",
file_type="CSV",
index4file="true",
breadth=2):
# Configurable parameters
self.planner_provider = planner_provider
self.planner_model = planner_model
self.writer_provider = writer_provider
self.writer_model = writer_model
self.file_type = file_type
self.index4file = index4file
self.breadth = breadth
file_type:
The type of the original input file. Currently, only CSV files are supported.
index4file:
Indicates whether the file includes an index column (typically the first column).
Accepts "true" or "false" as string values.
breadth:
Defines how many agents to invoke in parallel.
Each agent will generate three ideas. Ideally resulting in three tools if the code passes LLM-based evaluation.