Arthur is a platform for discovering, monitoring and governing AI agents. It finds agents across endpoints, cloud and private data centres, including unregistered shadow agents, and collects them in one inventory. It also provides step-level traces of reasoning and tool calls, evaluations covering for example groundedness and PII leakage, and runtime guardrails with policies per use case. Alerts can be streamed to SOC tools. According to its documentation, Arthur Shield, the firewall for LLM applications, is delivered either as single-tenant SaaS or as a self-hosted deployment.
Arthur is a US vendor and the platform is not open source. In the single-tenant SaaS option each customer runs in a dedicated AWS account and VPC in an AWS region of its choice; Azure or GCP requires a custom deployment. The customer can also keep the data in its own Postgres database. For a European organisation, the chosen region and the choice of customer-hosted data therefore determine where data resides, while the vendor itself remains subject to US jurisdiction.