Roark

Roark is a testing, monitoring and evaluation platform for voice AI agents. It is positioned as a QA and evals layer for agents a team already runs, not as an agent builder, and it works with Vapi, Retell, LiveKit, Pipecat and other stacks. Before launch, Roark simulates an agent against hundreds of callers, including personas, accents in 45 languages, background noise, adversarial red teaming such as prompt injection and jailbreaks, load tests and always on health checks. Suites can be rerun on every change and run in CI. In production, every call is scored on more than 500 audio native metrics such as pronunciation, empathy, resolution and compliance, with call replay, dashboards and OpenTelemetry traces. Problems are filed as issues, fixes are proposed as diffs, and Roark verifies on live calls that the metric moved. It offers a REST API, an MCP server and a CLI, and the vendor lists SOC 2 Type II and a HIPAA BAA. Pricing is usage based. Pay as you go starts at $0 with $50 of free credit and no card, with simulation at $0.15 per minute and metric evaluation at $0.04 per metric per minute, plus provider costs. Team is $500 per month spent as usage, at $0.10 per minute and $0.02 per metric. Enterprise starts at $4,000 per month committed, from $0.05 per minute, and adds SSO, custom roles and an uptime SLA.