WhoDunnitAI: Voice-Driven AI Murder Mystery Game

WhoDunnitAI enables real-time voice interrogation of AI suspects

WhoDunnitAI is a voice-driven murder mystery game where players solve a crime—specifically "Death at Blackwood Manor"—by interviewing five AI-powered suspects. The game leverages Large Language Models (LLMs) and voice synthesis to create an immersive experience where the player acts as a detective, asking arbitrary questions to uncover the truth behind a poisoned patriarch.

Technical Implementation and User Experience

The game utilizes a combination of speech-to-text (STT), LLMs for character persona management, and text-to-speech (TTS) for audio output. To manage costs and stability, the developer implemented several technical adjustments following the initial launch:

API Management: The developer introduced a "Bring Your Own Key" (BYOK) system to mitigate high operational costs. User-provided OpenAI keys are stored in the browser's localStorage and sent via the x-openai-key header to mint ephemeral client secrets for WebRTC functions.
Session Limits: To maintain service availability within a $500 budget, game session times were reduced to 15 minutes.
Infrastructure: The game uses WebRTC for real-time voice communication, though early users reported connectivity issues and "voice connection failed" errors during high-traffic periods.

Community Feedback and Analysis

User reactions to WhoDunnitAI on Hacker News highlight both the potential of voice-driven AI gaming and the persistent challenges of the medium.

Strengths and Innovation

Many users praised the concept as a long-awaited evolution of the adventure game genre. One user noted that this level of voice interaction has been a "dream of adventure game creators for so long," while others found the experience hooking even without a prior interest in crime mysteries.

Critical Pain Points

Critics and early adopters identified several areas for improvement:

AI Voice Quality: Some users expressed a strong aversion to AI-generated voices, suggesting that human narration would significantly improve immersion.
Narrative Consistency: A common concern in AI-driven mysteries is the tendency for models to hallucinate. One user shared an experience with a similar game where the AI suggested suspects were blood-related despite the game's established lore.
User Friction: The requirement to create an account acted as a barrier for some potential players, and others requested a demo video to evaluate the experience before signing up.
Game Design: Some players found the "vibe-coded" nature of the game to be bland or uninspired, while others reported bugs, such as the briefing timer remaining at 0:00.

Comparative Perspectives

Developers working on similar projects have noted that maintaining immersion in "social virtual reality" is difficult because a single contradiction can break the experience. One developer mentioned a complex pipeline involving over 40 separate steps to ensure narrative consistency before a prompt ever reaches the real-time API.