Arthur

Arthur is software for discovering, observing, and governing AI agents across enterprise environments. It monitors OpenTelemetry streams, MCP servers, network traffic, cloud APIs, and employee endpoints to find agents. For observability, it traces agent steps, tool calls, retrieved context, costs, latency, and behavior over time. Its governance tools support machine-checkable policies and real-time guardrail verdicts for prompt injection, PII exposure, restricted models, and off-policy tool use.

Deployment options include hosted SaaS, on-premises, and hybrid arrangements. In hybrid deployments, the evaluation engine runs alongside customer workloads so sensitive data remains in the environment. Arthur lists RBAC, SSO, and deployed engineering support with each deployment option. The Free plan is free and includes up to 4 use cases, 2 projects, and 7-day data retention. Premium is $60 a month. Enterprise pricing is custom. Platforms listed are API, self-hosted, and web.

Arthur combines agent discovery, observability, and governance with hosted, on-premises, and hybrid deployment choices. The Free plan is a defined starting option, while Premium is $60 a month and Enterprise pricing is custom.

Arthur suits enterprise teams that need to find and monitor AI agents, apply governance policies, or manage agent deployments alongside sensitive workloads. Its Free plan may suit teams with up to 4 use cases and 2 projects.

Overview

Arthur helps organizations find AI agents across enterprise systems, follow their activity, and apply operating rules. It is best suited to teams that need agent oversight across multiple environments, especially where deployment flexibility or data control matters. The combination of discovery, observability, and governance is its strongest case; the main trade-off is that project, usage, and retention limits shape how far the lower plans can stretch.

Key features

Discovery sensors monitor OpenTelemetry streams, MCP servers, network traffic, cloud APIs, and employee endpoints. That breadth suits organizations whose agents run across different systems, rather than in one framework alone. The range is valuable when it matches the environments a team needs to cover.

Arthur traces agent steps, tool calls, retrieved context, costs, latency, and behavior over time. Drift monitoring, model performance metrics, and bias monitoring add ways to follow changes beyond individual runs. Alerts can go to Slack or a webhook, giving teams two routes to bring findings into their existing workflows.

Machine-checkable policies and real-time guardrail verdicts address prompt injection, PII exposure, restricted models, and off-policy tool use. That makes Arthur a fit for teams that need controls during agent activity as well as a record of it. Its evaluation engine is open source and supports self-serve deployment, with built-in rules for PII, sensitive data, custom LLMs, and regex patterns.

Deployment options include hosted SaaS, on-premises, and hybrid setups. In hybrid deployments, the evals engine runs next to customer workloads so sensitive data can remain in the environment. RBAC, SSO, and deployed engineering support come with every deployment option, which makes the deployment choice less of a trade-off in those controls.

Documented integrations span OpenAI, Anthropic, LangChain, LiteLLM, CrewAI, LlamaIndex, OpenAI Agents, Google ADK, AWS Bedrock, and Mastra. Data connectors include BigQuery, Google Cloud Storage, Amazon S3, and Arthur Shield instances. This makes Arthur relevant to teams using those tools and stores; the documented set is the clearest basis for judging fit.

Pricing

The Free plan costs nothing (billed $0/mo). It allows up to 4 use cases, 1 organization, 1 workspace, and 2 projects, with 7-day data retention, 5k jobs, 300k spans, 12k inferences, and 3k evals. Unlimited seats make it accessible to a team, but the short retention and modest project and usage caps make it better for a small rollout than sustained, broad monitoring.

Premium costs $60 a month (billed $60/mo). It raises the allowances to 100 use cases, 10 projects, and 30-day retention, with 20k jobs, 1.2M spans, 100k inferences, and 75k evals; it still includes 1 organization and 1 workspace. This is the practical step up for teams that have outgrown the Free caps but can work within a single organization and workspace.

Enterprise has custom pricing. It includes unlimited organizations, workspaces, projects, and data retention, with custom jobs, spans, inferences, and evals, plus dedicated and managed VPC options. SSO, SLAs, and BAA are included, along with a dedicated customer success manager and advanced monitoring. It is aimed at organizations that need higher or tailored capacity and those enterprise support and deployment terms.

Platforms

Arthur supports API, self-hosted, and web access, with hosted SaaS, on-premises, and hybrid deployment choices. That spread serves teams that want a managed service as well as those that need workloads or sensitive data to stay within their environment.

Who it's for

Arthur suits organizations that need to discover agents across enterprise environments and govern their behavior, particularly when they use several of its documented integrations or require on-premises or hybrid deployment. Teams that only need a small, low-volume monitoring setup may find the Free plan sufficient. Those needing longer retention, more projects, or larger quotas should consider Premium; organizations requiring custom capacity, managed VPC, or enterprise terms are better matched to Enterprise.

Pros and cons

• Pro: Discovery spans telemetry, MCP, network, cloud, and endpoint sources, supporting oversight across varied enterprise environments.
• Pro: Tracing, monitoring, and real-time policy verdicts combine investigation with controls for specific agent risks.
• Pro: Hosted, on-premises, and hybrid deployment options accommodate different data and infrastructure needs.
• Pro: Free includes unlimited seats, so a team can share a small deployment without a seat cap.
• Con: Free retains data for 7 days and caps projects at 2, limiting its usefulness for longer investigations or broader rollouts.
• Con: Premium still limits customers to 1 organization and 1 workspace, so it does not fit every multi-unit deployment.
• Con: Enterprise pricing is custom, and its job, span, inference, and eval allowances are also custom, making capacity less predictable from the plan alone.

Alternatives

Machine Learning Model Monitoring Software and Model Monitoring Software are broader starting points for comparing monitoring tools.

Amazon SageMaker Autopilot is an alternative for readers who prefer pay-as-you-go pricing and do not need a free plan.

Deepchecks offers a free plan for up to 3 seats, 1 AI application, and 5K DPUs per month, with 3 months of data retention and unlimited prompt-based metrics. It suits readers whose needs fit those limits.

Evidently AI is an alternative with a fully open-source framework under the Apache 2.0 license.

NannyML offers a self-managed free Open Source plan and a Starter plan at $399 a month (billed $399/month), with 2 models and 10 M predictions. Readers comparing those plan terms can consider it alongside Arthur.

Opik makes its core observability and evaluation features available as open-source code that users can download and run locally.

Arize AX has a free SaaS plan with 25k trace spans per month, 1 GB ingestion per month, 15-day retention, and unlimited users and evals. It is worth considering for readers whose needs fit those caps.

Fiddler AI offers a free plan with real-time guardrails and latency under 80ms.

Galileo offers a Pro plan at $100 a month, billed yearly, with 50,000 traces per month and standard RBAC. Readers comparing a paid yearly plan with those limits may want to consider it.

Verdict

Choose Arthur if your organization needs to discover agents across multiple enterprise systems, inspect their behavior, and apply real-time governance, with the option to keep workloads in your own environment. Its central advantage is bringing those functions together across several deployment choices. Look elsewhere if your needs exceed its lower-plan quotas but you do not need Enterprise’s custom capacity and terms, or if you want an alternative whose open-source or plan limits better match your use case.