Evalab’s Ghost Services for AI Agents

Field notes

The displayed workflow redirects an agent’s service endpoints to isolated twins, then records each provider call, response, latency, side effect, and service-state change. The sample environment serves Slack, GitHub, and Google Calendar endpoints and reports four provider calls in 266ms, with a GitHub check-run blocked. The page also describes short-lived runs that expire after 60 minutes, alongside permanent sessions and scenario seeding from mock or production data.