Dioptra is an open-source software test platform from NIST for assessing the trustworthy characteristics of artificial intelligence systems. Published on GitHub with about 300 stars and actively updated in October 2026, it supports the Measure function of the NIST AI Risk Management Framework.
Trustworthy AI, in NIST’s framing, means systems that are valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed. Dioptra gives evaluators a platform to assess, analyze, and track those characteristics through designed experiments. It provides a REST API that can be driven from an intuitive web interface, a Python client, or any REST client library, for designing, managing, executing, and tracking experiments. Experiments are composed from modular components described in simple YAML files, a plugin system allows importing existing Python packages, and the platform automatically snapshots resources so experiments can be reproduced and validated, with full input history tracked for traceability.
Dioptra’s stated design goals include reproducibility, traceability, extensibility, interoperability, modularity, and security, and the software is at release 1.2.1 with ongoing development. Its use cases span first-party model testing during development, second-party assessment during acquisition or in an evaluation lab, and third-party auditing and compliance activities, as well as red-teaming in a controlled environment.
Dioptra is free open-source software from the National Institute of Standards and Technology. There is no paid edition or hosted service; evaluation teams run the platform in their own environment.
Dioptra is for AI evaluation labs, auditors, government teams, and model developers who need structured, reproducible evidence about how a model behaves before it is acquired or deployed.