KAGIDOKO : A Deep Learning Horror Game (v0.4.05)

A hostile house turns hide-and-seek horror on its head by making hiding impossible. The threat stalking its corridors is a ghost controlled by a compact deep-learning model, and survival depends less on finding a closet than reading its behavior before it reaches you. Every hallway feels like a trap, every dead end becomes a gamble, and the mood stays tense because the creature can return even after you think you have dealt with it. KAGIDOKO : A Deep Learning Horror Game is a short, focused nightmare built around observation, prediction, and the dread of being caught unaware.

The goal is straightforward: search the house for keys, unlock its sealed doors, and find a way out without suffering a jumpscare. Instead of fleeing from the ghost, you use light to confront it directly; spotting the entity and keeping it in view under illumination forces it to vanish. That simple reversal creates the central loop: move through unfamiliar passages, listen and watch for danger, anticipate where the ghost may be lurking, then turn the tables before it attacks from behind. Its neural-network-driven behavior aims to feel less like a monster following a fixed patrol and more like a small, reactive creature making moment-to-moment decisions.

What stands out is the unusual use of machine learning as an actual gameplay ingredient rather than a buzzword pasted onto a conventional scare ride. The game’s “mosquito-like” AI concept gives the haunting a deliberately unpredictable edge, turning each search for keys into a test of awareness and nerve. The stripped-down structure also keeps the focus on its single clever idea instead of burying it beneath elaborate lore or inventory systems. Players who enjoy the helpless corridor tension of indie chase horror, but want a more active answer than simply running and hiding, should find its light-based confrontations especially appealing; veterans may appreciate its experimental premise most.