Expansive Library of
Guardrail Metrics
Choose from general-purpose metrics or create custom metrics tailored to your domain—each one precisely detecting whether hallucinations exist in your AI outputs with granular 0-100 scores.
Evaluates whether each claim in the AI output is true and verifiable—critical for healthcare, finance, and legal domains.
Assesses whether the response addresses all parts of the prompt with sufficient detail and relevance.
Checks whether the AI followed explicit instructions including tone, structure, format, and style guidance.
Determines whether each factual claim is directly supported by the provided context—essential for RAG systems.
Measures how closely the output matches a provided correct answer—ideal for benchmarking and fine-tuning.
Detects PII, hate speech, self-harm content, prompt injections, CBRN, and other safety violations.
Evaluates how effectively the AI autonomously plans, decides, and executes complex multi-step tasks.
Validates that AI tool calls are accurate and properly formatted—essential for agentic workflows.
We're building more general-purpose metrics to help you evaluate and improve AI outputs across a wider range of use cases.
Register your own prompt-based metric or purchase a fully built, guaranteed metric from our engineers—both available on Pro and Enterprise plans.
Add domain-specific checks instantly with your evaluation and fix prompts. Perfect for rapid experimentation across Detect, Score & Fix flows.
Partner with DeepRails engineers to design, test, and stand up a production-grade metric with accuracy guarantees—ideal for regulated or high-stakes workloads.
Test every metric free
in the Playground
Zero code, zero setup. Run your own outputs through any guardrail and watch it detect, score, and fix.