Arthur

Arthur is an AI governance and monitoring platform offering one control plane to discover, observe, evaluate, and secure AI models and agents. It provides step-level agent observability, continuous evaluations, runtime guardrails and behavioral analytics, plus cost tracking and model rerouting. It maintains an open-source Arthur Engine for self-hosted evaluation and supports SaaS, VPC, and on-prem deployment. Plans start free, with a $60/mo Premium tier and custom Enterprise pricing.

Arthur (Arthur AI) provides a centralized control plane for trustworthy enterprise AI, spanning model monitoring, observability, evaluations, guardrails, and governance across the AI lifecycle. Its agent discovery finds every agent running across endpoints, cloud, and on-prem environments (including shadow, unregistered agents), then gives full step-level observability into each agent's reasoning, tool calls, retrieval, and handoffs so teams can see not just what an agent did but why. For runtime protection, Arthur's Agent Behavioral Analytics continuously monitors AI applications and alerts on anomalies such as prompt injection, anomalous tool sequences, and data-egress spikes, streaming findings to SOC tools like CrowdStrike Falcon, Elastic, Splunk, and Datadog. It also offers cost controls that track spend by team and can reroute requests that don't need an expensive model to cheaper or internally hosted open-source models. Arthur maintains an open-source Arthur Engine on GitHub for self-serve, self-hosted evaluation. Arthur is used by enterprise AI, security, audit, and governance teams and offers deployment as multi-tenant SaaS, single-tenant SaaS, and self-managed VPC/BYOCloud/on-prem. The honest limitation: it is a technical, ML/security-team-oriented platform (not a business-user tool), the richest deployment, SSO, SLA, and BAA options are gated to the custom-priced Enterprise tier, and the free and Premium tiers cap use cases, data retention, and volume.