The TrustyAI Kubernetes Operator is a Go-based Kubernetes operator designed to simplify the deployment and management of TrustyAI components within Kubernetes and OpenShift environments. The operator serves as a cloud-native solution for orchestrating explainable AI and fairness monitoring infrastructure, addressing the need for automated lifecycle management of AI services in containerized environments.
The operator manages three primary TrustyAI components. The TrustyAI Service deploys alongside KServe models to collect inference data, enabling model explainability, fairness monitoring, and drift tracking capabilities. FMS-Guardrails provides a modular framework for guardrailing large language models. LM-Eval implements a job-based architecture for deploying and managing LLM evaluations, built on EleutherAI's lm-evaluation-harness library. This multi-component approach allows organizations to implement comprehensive model monitoring and explainability across their ML infrastructure.
The operator requires Kubernetes v1.19 or later, or OpenShift v4.6 or later, with corresponding kubectl or oc client versions and kustomize v5 or higher. The operator is distributed as a container image on Quay.io, enabling straightforward deployment on existing clusters. The project maintains automated testing through controller tests, YAML linting, and Gosec security scanning workflows, with Go Report Card integration for code quality tracking.
The project/lm-eval label appears on 22 items, reflecting significant ongoing work related to LLM evaluation functionality.
The project is classified across multiple domains including AI Explainability, Kubernetes Operator, Trustworthy AI, ML Models, Service Deployment, Model Monitoring, Bias Detection, AI Fairness, Cloud Native, and Explainable AI, reflecting its multifaceted role in the ML operations landscape.
The operator is licensed under Apache License Version 2.0 and maintains contribution guidelines documented in a CONTRIBUTING.md file. Documentation is available through the OpenDataHub project, which provides configuration guidance for TrustyAI monitoring. The project participates in Hacktoberfest, indicating openness to community contributions. The combination of automated testing, security scanning, and code quality monitoring demonstrates a commitment to production-ready operator standards for managing AI explainability and fairness infrastructure in Kubernetes environments.
