OpenVals
1. Start here

Your OpenVals roadmap

OpenVals is the trust layer for AI. Use this docs set as a map: start with why it exists, run your first evaluation, learn the trust metrics, then move into platform workflows, integrations, and enterprise readiness.

The question OpenVals answers
Can you trust this model for this use case?

What OpenVals does

Traditional benchmarks stop at accuracy. OpenVals measures the signals that matter before production: correctness, factuality, hallucination risk, reliability, safety, latency, dataset quality, and decision readiness.

Evaluate models before deploy
Benchmark across labs and sizes
Score trust with DRS / HPI
Validate datasets, not just models
Generate executive evidence packs
Guide use-case and readiness decisions

The learning path

Follow these steps in order the first time. Jump ahead later once you know which layer you need.

01
Understand the problem OpenVals solves

See why demo accuracy fails in production, and what trust questions every team should ask.

Read Why OpenVals
02
Install the CLI / SDK

Get OpenVals on your machine so you can run evaluations locally and inspect your environment.

Install OpenVals
03
Run your first evaluation

Benchmark a model, validate a dataset, and produce a trust report in minutes.

Start first evaluation
04
Learn the trust metrics

DRS, factuality, hallucination probability, reliability, and how to interpret a good vs risky score.

Open DRS guide
05
Compare models and domains

Run multi-model benchmarks across finance, software, cybersecurity, healthcare, and more.

Explore benchmarking
06
Use the platform workflows

Connect providers, run Use Case Advisor, launch validation runs, and read evidence packs.

Open platform guide
07
Assess enterprise readiness

Use AIxLence when you need capability diagnosis, roadmap sequencing, and board-ready outputs.

Open AIxLence docs

Pick your starting lane

What you will learn to measure

AreaWhat you getGo to
Trust scoreDecision Reliability Score across key signalsDRS
CorrectnessModel evaluation and factuality checksModel evaluation
Hallucination riskProbability bands and failure modesHPI
Data qualitySchema, balance, and dataset healthDataset intelligence
ComparisonMulti-model ranking on real domainsBenchmarking
EvidenceExecutive reports and UI previewsReporting

Go deeper when you need it

  • API contracts and architecture → Platform API Spec and Architecture Flow
  • Domain datasets → Supported Benchmark Domains
  • Security / contribution norms → Security Policy and Contributing
  • Product direction → Product roadmap and Vision

Prefer a guided first hour? Start with Why OpenVals, then Install, then First evaluation.