An AI evaluation, testing, and governance platform founded in San Francisco in 2021, which raised a $14.5M Series A in May 2025 for roughly $19.3M total funding with a team of 23–28. It spans seven capability layers: discovery (AI system inventory plus Shadow AI detection), registration (risk tiering and lifecycle management), testing (175+ built-in evaluations with CI/CD regression blocking), observability (input/output tracing with quality, cost, latency, and drift monitoring), guardrails (real-time blocking of prompt injection, PII/PHI leakage, and jailbreaks), compliance (mapping to EU AI Act, NIST AI RMF, and ISO 42001), and cost control (token tracking by project/team/user with hard budget caps). The free Basic tier includes 1 member, 5 projects, 20,000 inferences per month, CLI/SDK/API access, observability, and alerts. Enterprise is custom and adds SSO, RBAC, on-premise deployment, SOC 2 Type II reports, and a 99.99% SLA. Verification includes a G2 rating of 4.3/5 from 18+ reviews, Product Hunt at 5/5 from 6+ reviews, a named KPN (Dutch telecom) deployment where the AI Tech Lead publicly described scaling their eval-driven philosophy, and a ChatGPT plugin for viewing projects, test results, and governance controls.
Pros & Cons:
Pros: Covers the full AI lifecycle—inventory, testing, observability, guardrails, compliance mapping, and cost control—rather than selling a single layer, which reduces tool sprawl for teams building governance programs; 175+ built-in evaluations with CI/CD regression blocking turn testing into an enforceable gate rather than a one-off check; guardrails block prompt injection, PII/PHI leakage, and jailbreaks in real time, complementing the evaluation layer; compliance mapping to EU AI Act, NIST AI RMF, and ISO 42001 addresses a real and growing reporting burden; a free tier with 20,000 inferences per month and CLI/SDK/API access supports genuine evaluation; the KPN deployment is a named, on-the-record customer reference rather than an anonymous logo.
Cons: The breadth of seven layers can mislead buyers—teams looking for a single-purpose guardrail or testing tool may not realize they are evaluating a governance platform, so positioning must be stated clearly; Enterprise pricing is not public, and the free tier’s 1 member / 5 projects limit means production use requires a sales conversation; independent review volume is moderate—G2 has 18+ reviews, far fewer than established helpdesk or video API vendors, so confidence rests on a smaller sample; guardrails are a 2026 expansion layer, not the company’s historical core, which was evaluation and observability, so it should not be described as a native security vendor; name collision with the open-source OpenLayers JavaScript mapping library creates search confusion and requires explicit disambiguation.
Who is it for?:Engineering teams building AI features that need evaluation, observability, and regression gates in CI/CD; compliance and risk teams mapping AI systems to EU AI Act, NIST AI RMF, or ISO 42001; organizations that need both testing and runtime guardrails in one platform rather than stitching vendors; teams that want a free tier with real inference volume before committing to Enterprise.