Overview

Arthur brings agent discovery, observability, and governance together for organizations operating agents across enterprise systems. It is best suited to teams that need visibility across multiple environments and want controls during agent activity, not just a record after the fact. Its broad monitoring and deployment choices are a strong fit for enterprise requirements, though the lower plans impose firm project, retention, and usage caps.

Key features

Discovery sensors monitor OpenTelemetry streams, MCP servers, network traffic, cloud APIs, and employee endpoints. That range can help teams find agents beyond a single framework or cloud, provided the sensors cover the systems they need to manage. Arthur then traces agent steps, tool calls, retrieved context, costs, latency, and behavior over time, giving teams material to investigate outcomes and track changes. Drift, model performance, and bias monitoring add signals beyond individual traces; Slack and webhook alerts can route notifications into existing workflows.

Governance pairs machine-checkable policies with real-time verdicts for prompt injection, PII exposure, restricted models, and off-policy tool use. That makes Arthur relevant to teams that need to intervene against specific risks while agents run, rather than relying only on post-run analysis. Documented integrations span OpenAI, Anthropic, LangChain, LiteLLM, CrewAI, LlamaIndex, OpenAI Agents, Google ADK, AWS Bedrock, and Mastra. Platform connectors include BigQuery, Google Cloud Storage, Amazon S3, and Arthur Shield instances.

Arthur supports hosted SaaS, on-premises, and hybrid deployment. Its hybrid evals engine runs alongside customer workloads so sensitive data can stay in the environment. RBAC, SSO, and deployed engineering support come with every deployment option, a meaningful baseline for teams with access-control and implementation needs. The open-source Evals Engine supports self-serve deployment and includes PII, sensitive-data, custom-LLM, and regex rules.

Pricing

The Free plan costs 0.00 USD per month, billed $0/mo. It includes up to 4 use cases, 1 organization, 1 workspace, 2 projects, 7-day data retention, 5k jobs, 300k spans, 12k inferences, and 3k evals. Unlimited seats lower the barrier for a team evaluation, but the short retention and small project allowance make it a poor fit for sustained, broad monitoring.

Premium costs 60.00 USD per month, billed $60/mo. It raises the caps to 100 use cases, 10 projects, 30-day retention, 20k jobs, 1.2M spans, 100k inferences, and 75k evals, while retaining the 1-organization and 1-workspace limits. It is the practical step up for teams that have outgrown the Free quotas, though organizations needing multiple workspaces or longer retention will need another arrangement.

Enterprise has custom pricing and offers unlimited organizations, workspaces, projects, and data retention, with custom job, span, inference, and eval volumes. Dedicated and managed VPC options, SSO, SLAs, BAA, advanced monitoring, and a dedicated customer success manager suit organizations with larger operational or compliance needs. Choose it when Premium’s fixed scope is too restrictive and those enterprise provisions justify custom pricing.

Platforms

Arthur is available through API and web, with self-hosted deployment also supported. The mix of SaaS, on-premises, and hybrid options makes it more adaptable to data-location constraints than a hosted-only setup; the hybrid evals engine is particularly relevant when sensitive data must remain alongside customer workloads.

Who it's for

Arthur is a strong candidate for teams responsible for agent fleets spread across enterprise systems, especially where tracing, runtime policy checks, and deployment flexibility belong in one operating layer. It is less compelling for a small project that needs only basic evaluation or has volumes beyond the Free limits but cannot justify Premium or custom Enterprise pricing.

Pros and cons

Pro: Discovery spans streams, servers, traffic, cloud APIs, and employee endpoints, helping teams look beyond agents built in one framework.
Pro: Real-time guardrail verdicts and machine-checkable policies address concrete risks such as PII exposure and off-policy tool use.
Pro: Hosted, on-premises, and hybrid deployment options support different data-control requirements, with RBAC, SSO, and deployed engineering support across options.
Con: Free retention is 7 days and limited to 2 projects, so it is suited to evaluation rather than broad ongoing coverage.
Con: Premium remains limited to one organization and one workspace, despite higher project and volume caps.

Alternatives

For teams prioritizing model monitoring rather than Arthur’s combined agent discovery and governance, compare Machine Learning Model Monitoring Software and Model Monitoring Software.

Amazon SageMaker Autopilot is worth considering when pay-as-you-go pricing and a free trial suit the workload better; its Autopilot-specific rate is not given.
Deepchecks fits teams seeking a freemium option with a self-hosted platform and a Basic tier capped at 3 seats and 1 AI application.
Evidently AI is a fit for teams that prioritize a fully open-source Apache 2.0 framework.
NannyML offers a self-managed open-source plan and a Starter plan at 399.00 USD per month for teams needing a different monitoring option.
Opik suits teams that want an open-source core observability and evaluation feature set they can download and run locally.
Arize AX may suit a smaller SaaS evaluation, with its free tier capped at 25k trace spans per month and 15-day retention.
Arize Phoenix is a free, open-source, self-hosted alternative for teams that want user-managed trace spans, ingestion, and projects.
Fiddler AI offers a free tier with real-time guardrails and latency under 80ms, a different focus for teams comparing guardrail options.

Verdict

Choose Arthur if your team needs to discover agents across enterprise environments, inspect their behavior, and apply runtime policies without committing to one deployment model. Its breadth and hybrid option are the strongest reasons to adopt it; the Free plan’s short retention and Premium’s single-workspace ceiling are the clearest reasons to look elsewhere or move to custom-priced Enterprise.

	Arthur	STAMM
Free plan	Yes
Paid from	$60/mo
Platforms	api, self-hosted, Web	api, Linux, macOS, self-hosted, Web, Windows
Free plan	Yes
Paid from	60 /mo
Drift monitoring	Yes	Yes
Model performance metrics	Yes	Yes
Bias monitoring	Yes
Alert channels	webhook, Slack
Deployment options		self-hosted

Arthur vs STAMM

Arthur

STAMM