Arthur has open-sourced the Arthur Engine, a real-time evaluation and guardrail tool for generative AI and traditional machine learning models, the company said on 31 March 2025.

The engine runs locally, which Arthur says preserves data sovereignty and keeps all data inside the customer’s infrastructure.

Features include real-time evaluation to detect failures before they reach production, active guardrails to prevent hallucinations and bad outputs, customisable metrics, and support for GPT, Claude, Gemini, open-weights models and traditional ML.

Arthur cites risks such as data leaks, model drift and slow debugging cycles as reasons teams need ongoing monitoring.

“Open-sourcing the Arthur Engine puts powerful AI evaluation tools into the hands of developers, researchers, and builders worldwide,” said Ashley Nader, Lead AI PM at Arthur.


Share this post
The link has been copied!