API Reference

Generated from docstrings in src/uaef/. If something here looks wrong, fix the docstring.

Core

Module	What it holds
uaef.models	AgentTrace, Message, ToolCall, GroundTruth, MetricScore, EvaluationResult
uaef.adapters	Framework adapters and the adapter registry
uaef.metrics	All built-in metrics, BaseMetric, the metric registry
uaef.evaluation	evaluate() and the evaluator classes

Advanced

Module	What it holds
uaef.llm_judge	Bedrock judge, prompt templates, calibration
uaef.experiments	Experiment management, comparison, regression detection
uaef.analysis	Root cause, clustering, trends, recommendations
uaef.reporting	Dashboards, deep dives, safety reports, export

Specialized

Module	What it holds
uaef.hitl	Human-in-the-loop review queues, agreement, calibration
uaef.security	PII detection and redaction, trace encryption
uaef.storage	DynamoDB + S3 persistence
uaef.client	UAEFClient remote SDK
uaef.integrations	RAGAS and DeepEval connectors
uaef.simulator	Conversation simulation — placeholder, not implemented

Common entry point

Most callers need only evaluate():

from uaef.api import evaluate

result = evaluate(
    trace=trace,
    ground_truth=ground_truth,
    metrics=["answer_relevance", "safety_score"],
    persist=True,
    experiment_name="Customer Support Agent v2",
)

Worked examples live in the Guides.