Advanced‐Features

Advanced Features

Confidence-Aware Routing

In production, you don't always want to trust a model prediction if the probability is low.

Example Pattern:

tool = ModelTool(
    model=clf,
    confidence_threshold=0.85,
    on_low_confidence="fallback",
    fallback_tool=human_review_tool
)

Model Ensembles You can wrap multiple models into a single "Super Tool."

Mean Strategy: Great for regression (averaging house prices from 3 different models).

Vote Strategy: Great for classification (if 2 out of 3 models say "Churn", the tool returns "Churn").

This abstracts the complexity away from the LLM; the Agent thinks it is talking to one expert, but it is actually getting the consensus of an entire ensemble.