Skip to main content
TrustifAI gives you a principled, multi-dimensional Trust Score for any LLM or RAG response — going far beyond a single correctness check. Instead of a black-box number, TrustifAI breaks trustworthiness into four orthogonal signals (evidence coverage, epistemic consistency, semantic drift, and source diversity), combines them with configurable weights, and renders an interactive reasoning graph that shows exactly why a response was deemed reliable or unreliable.

Quickstart

Score your first RAG response in under five minutes with a working code example.

Installation

Install TrustifAI via pip and configure your environment variables.

Core Concepts

Understand how Trust Score is computed from four independent trust signals.

API Reference

Explore the full public API: Trustifai, AsyncTrustifai, MetricContext, and more.

​
What TrustifAI evaluates

TrustifAI computes trust across two evaluation modes: Offline metrics — for already-generated RAG responses: Online metrics — for real-time generation:

​
Why TrustifAI

Most evaluation frameworks return a single scalar. TrustifAI returns a structured, explainable result:
  • Weighted aggregation — tune metric importance to your use case (e.g., weight evidence coverage higher for medical Q&A)
  • Reasoning graph — a DAG visualization showing metric scores, aggregation, and the final decision
  • Custom metrics — plug in your own evaluation logic without touching core library code
  • Async-first — native async pipeline with concurrency control and rate limiting for large-scale batch evaluation
  • LiteLLM-backed — works with OpenAI, Anthropic, Gemini, Azure, Mistral, Ollama, and more

Offline metrics explained

Deep dive into Evidence Coverage, Epistemic Consistency, Semantic Drift, and Source Diversity.

Online metrics explained

Learn how the Confidence Score uses token log probabilities for real-time trust signals.

Reasoning graphs

Understand and customize the interactive DAG visualization of your evaluation logic.

Custom metrics

Extend TrustifAI with your own evaluation logic using the BaseMetric interface.