confidence-router · packages

Evaluates classification results against configurable confidence thresholds to determine whether to route to a specific target, request clarification, or trigger a fallback. It provides a `ConfidenceRouter` class that acts as a decision engine for classification pipelines.

Provides a collection of intent classification strategies—including keyword matching, embedding similarity, and LLM-based analysis—that implement a unified `Classifier` interface for use with `confidence-router`. It includes a `ClassifierRegistry` class to manage these implementations and execute them in configurable fallback chains.

Provides the core type definitions, configuration utilities, and the `DecisionEngine` class for evaluating classification confidence against routing thresholds. It serves as the foundational library for building custom routing logic within the confidence-router ecosystem.

Optimizes `routeThreshold` and `fallbackThreshold` values for any `RouterInterface`-compliant router by performing a grid search against a labeled dataset to maximize F1 scores. It provides a `ThresholdOptimizer` class that returns detailed performance metrics, including accuracy, precision, recall, and confusion matrices.

Provides localized clarification prompts and locale-aware formatting for the `confidence-router` ecosystem. It exports `LanguageManager` and `PromptGenerator` classes to handle 47 built-in languages, RTL support, and custom language definitions.