Callcheck
About this project
In one sentence, what are you building?
CallCheck tests a Vapi voice agent, finds where it fails, proposes a prompt fix, and proves with before-and-after results, including a test it never saw, whether the fix worked.
What problem are you solving?
Teams that build voice agents have no reliable way to know where their agent fails, and no way to know whether a fix worked. Today a developer calls the agent a few times by hand, ships it, and hears about problems from angry callers. When they do spot a flaw, they edit the prompt and hope. Voice agents also fail inconsistently, as yours did, so one good test call proves nothing. Your demo shows the cost. The pizza agent told a caller with a severe nut allergy that the veggie pizza contains no nuts, when it has no ingredient information at all. In a real business, that kind of mistake can hurt someone, and nobody would find out until after it happened. CallCheck finds that failure with a real test, writes a targeted fix, applies it only with your approval, and then reruns the tests, including one the fix never saw, to show whether it really worked. The result is evidence, not a guess.