A developer has created a voice-driven murder mystery game where players interview AI suspects using speech-to-speech technology. The project uses OpenAI's gpt-realtime-2.1 over WebRTC for realistic conversations, though the model's high cost necessitates restrictions like authenticated user IDs and a 30-minute timer.

When players make direct accusations, each suspect triggers a tool that captures who was accused and the evidence presented. A separate gpt-5-mini judge evaluates whether players genuinely stated the required evidence facts, with paraphrasing counting but vague suspicion or fishing not accepted.

The game is built using Next.js, MongoDB, and Clerk for authentication. The developer originally built a prototype 2-3 years ago when AI was already popular but voice agents were still early, and has now revisited the technology to create a more refined version.

Players can test their interrogation skills against AI suspects to see if they can solve the mystery through real questioning.