TruLens

TruLens is an open-source library for evaluating and tracking LLM application quality, providing a structured workflow for measuring output quality, diagnosing failure modes, and monitoring performance over time in production. Its core abstraction is the Feedback Function, a configurable quality metric that evaluates a specific dimension of model output using an LLM judge, a trained classifier, or a heuristic rule. Pre-built feedback functions cover Groundedness (is the response supported by the context?), Answer Relevance, Context Relevance, Harmlessness, Conciseness, and Correctness for common RAG and chatbot use cases. The TruLens dashboard provides a visual interface for reviewing application traces, filtering by quality score, and identifying the specific inputs that produce the worst outputs, enabling targeted debugging of failure modes rather than reviewing individual examples randomly.

The Leaderboard view compares different application configurations (model version, prompt, retrieval strategy) across all metrics simultaneously, enabling data-driven selection of the optimal configuration. TruLens integrates with LangChain, LlamaIndex, and direct LLM API usage through a decorator-based instrumentation model that captures application traces without invasive code changes. Open-source and free. Hosted TruEra platform provides enterprise features for team collaboration and production scale evaluation pipelines.

Commonly used alongside Ragas and Langfuse for comprehensive LLM quality assurance in teams building RAG systems and conversational agents.