The AI Escape Room Quiz Powered by Emergent NPCs.
AI-powered escape room quiz game where non-player characters (NPCs) leverage LLM backends for emergent, unpredictable, and dynamic gameplay logic.
Active Gaming R&D
In DevelopmentZaqwes is currently in early prototype stages. We are researching autonomous NPC goal reasoning, anti-jailbreak game bounds, and generative riddle synthesis. Playtest slots will be announced to Discord members first.
No Scripted Dialogues. No Fixed Solutions.
Every escape room quiz in Zaqwes is generated with bespoke constraints. You don't click multiple-choice options — you interrogate, convince, bargain, and outsmart living LLM-backed actors.
1. Emergent NPC Minds
NPCs possess hidden memories, personal biases, and emotional triggers. Catch them in contradictions or appeal to their sympathies to reveal door ciphers.
2. Dynamic Escape Quizzes
Riddles and room mechanics adapt to your player style. If you are aggressive, security tightens. If you solve puzzles cleverly, the environment reacts dynamically.
3. High Stakes Timer Loop
Every question, dialogue turn, and failed attempt costs room oxygen and battery reserves. Rapid, clever deductions are critical to survive.
Hybrid Deterministic / Generative Engine
How do you stop LLMs from breaking escape room rules or hallucinating keys that don't exist? Zaqwes combines rigid deterministic state machines (locks, inventory, physics) with generative narrative models.
State Guardrails & Verification
NPC dialogue cannot hallucinate door unlocks unless the verification engine confirms mathematical conditions are met.
Dynamic Psychology State Machine
NPCs track suspicion, fear, and admiration metrics that directly color their token generation prompts in real time.
Zero Pre-Written Dialogues
Every interaction is authentically negotiated. Your creativity defines your escape path.
Explore Sibling Projects
Want to test the AI escape room first?
Join our Discord community. We run weekly closed playtests to push LLM NPC boundaries, balance dynamic quizzes, and refine prompt engines.