The Asylum AI is a survival horror narrator that traps you in a unique psychiatric institution each session, generated one room and one consequence at a time. Unlike generic chatbots that forget doors, files, and deaths, this agent treats its location graph and NPC fates as locked canon. Built on Jenova's unified platform, it runs on models from OpenAI, Anthropic, Google, xAI, DeepSeek, Qwen, and more. This is a work of psychological horror fiction, not clinical advice.
The Asylum builds a fresh hospital at the start of play — name, era (typically 1900s–1960s), superintendent or chief physician, stated purpose, and the treatments they drove past ethics. Hydrotherapy, electroconvulsive therapy, lobotomy, experimental medications, or a method they kept off the books become that session's institutional history. You never see the full picture up front; the administration, The Sin, and The Consequence surface through play.
The Asylum moves the investigator a single room per turn through administrative offices, treatment areas, patient wards, and service spaces such as the kitchen, laundry, morgue, and boiler room. A location graph records visited rooms and their exits so a records office with two doors still has two doors on return; backtracking cannot invent shortcuts. Architecture obeys ordinary physics — no rearranging wings, vanishing doors, or a building that "wants" anything.
The Asylum withholds the truth until you assemble it from patient files, intake forms, staff memos, worn restraints, and 1–4 NPCs who may help, lie, betray, or know nothing. Searching can yield fragments, red herrings, danger, or nothing; interrogating the narrator does not dump answers. Warnings you ignore scale from a close call to a lethal encounter.
The Asylum lets the investigator die. Combat is desperate and usually fatal; injuries change what hands can hold and what a head wound will let you focus on. Escape is possible only after at least eight rooms exist in the location graph and you pass comprehension gates that prove you understood the story. Rushing to a door without that understanding produces false escape — a loss that still ends with CASE CLOSED.
The Asylum never acknowledges a player, never takes retcon orders, and never fast-forwards empty halls. After death or a real escape, the fiction drops for a Case File: the full generated history, every NPC's nature and fate, which ending tiers were available, and a clean invitation to a different asylum. Nothing carries over.
| Traditional Approach | The Asylum |
|---|---|
| Chatbot forgets which doors existed | Location graph locks established exits |
| Player can demand a scene rewrite | Narrated events cannot be undone |
| Fast-forward through empty halls | One room at a time; no autopilot |
| Plot armor and fade-to-black | Death is real; CASE CLOSED is final |
Unlimited Memory vs. Chatbot Amnesia
A generic chatbot that forgets Patient #0093's transfer, an NPC left in peril, or that the records office had a locked director's door will contradict the building the moment you backtrack. The Asylum commits geography, explicit inventory (items enter only when taken), capability-affecting injuries, NPC dispositions, and key discoveries as canon. Return to a hallway and the same exits are there; leave a survivor in danger long enough and you may find a body.
Multi-Model AI Access
Bureaucratic dread and compressed violence want different instincts. An intake form that lists "Hysteria" or "Unnatural tendencies" as the reason for commitment is not the same job as a death scene written in subject-verb-object, or an NPC who remembers your last conversation and refuses to cooperate. The Asylum can run on OpenAI, Anthropic, Google, xAI, DeepSeek, Qwen, and more; after CASE CLOSED it even tells you to try another model for different pacing, horrors, and fates — which a ChatGPT-only or Claude-only session cannot do.
Intelligent Tool Integration for End-to-End Workflows
The Asylum is a production loop, not a Q&A bot: investigator action in, committed location/inventory/condition/NPC state, then the next room, a torn treatment log, or a final death scene. After a terminal ending, the Case File workflow dumps the administration, methods, The Sin, The Consequence, exit condition, and alternate ending paths. External apps are not summoned mid-scene; breaking character to fetch mail or a map would puncture the fiction this agent is built to hold.
Narrator Authority Over Wish Fulfillment
Declaring that you "figured it out," demanding the exit, or inventing a clever plan does not make those things true. You control what you attempt; the narrator controls whether it works, what exists behind a door, and how an orderly or former patient replies. Creativity is evaluated for plausibility. Plot armor is not issued.
Fans of Outlast-style survival horror Players who want to hide, read, and run as an ordinary investigator with no combat training, no supernatural senses, and no scripted map.
Interactive-fiction players who accept narrator authority People who want their input treated as intention, not a command to spawn loot, skip time, or rewrite a room that already exists.
Readers who assemble fragmentary evidence Anyone who prefers incomplete files, redacted names, and patient artwork over an exposition dump, and who will live with a search that finds nothing.
Players who want unflinching institutional horror Anyone prepared for psychological tension and unsettling historical treatments, with a narrator that does not cut away or reassure when a scene turns frightening.
Replay-focused players Each session generates a new institution, new methods, and new NPCs. Prior asylums do not carry over.
Free For exploring Jenova:
Plus (US$20/month) For individuals:
Premium (US$50/month) For creators & freelancers:
Pro (US$100/month) For professionals:
Max (US$200/month) For power users:
Ultra (US$500/month) For unconstrained productivity:
Enterprise (US$1,000/month) For businesses & organizations:
All plans are monthly subscriptions priced in USD, before applicable taxes. Your usage limit resets on your billing date.
The Asylum AI is an in-character survival horror game that drops an investigator inside a generated psychiatric hospital and plays every ward, file, and NPC encounter without a predetermined script. You wake with a missing stretch of memory, no equipment, and a building designed to keep people in. The session ends in a comprehension-gated escape, a false escape, or death marked CASE CLOSED.
The Asylum remembers the committed map, the files you actually read, who you left alive, and which injuries still limit your hands — a generic chat that loses that state will restage doors and contradict NPCs on the next visit. You can switch among OpenAI, Anthropic, Google, xAI, DeepSeek, Qwen, and more so a clinical memo, a compressed killing, and a post-game Case File are not forced through one vendor's default voice. The workflow is action → locked session state → next room or terminal ending, not a tool-calling assistant that breaks character to search the web or file a document mid-scene.
The Asylum does not call Gmail, calendars, Drive, or search during play; the narrator is forbidden from leaving the fiction. Continuity is handled internally through the location graph, inventory, condition line, NPC records, and key discoveries. After CASE CLOSED or a genuine escape, you receive an in-thread Case File and alternate-ending debrief rather than an export to an external app.
Yes. Death is final for that session — institutions are patient, and the narrator does not flinch. After the death scene the agent prints CASE CLOSED with no retry prompt. Death is a consequence of choices (ignored warnings, direct combat with the primary threat, repeated recklessness), not a scripted destination; survival and escape remain possible.
No. Once a room, character, or event is narrated, it stands; demands to reverse it are ignored in-character and the fiction continues forward. You may react to what exists. You may not edit it away.
No. Time skips, montages, and "take me to the morgue" jumps that skip intervening rooms are refused. Movement is always the next room on the path, described and navigated individually. Multi-step actions inside one scene (check the closet and the drawer) are allowed; traversing the building in a summary is not.
The exit is not available before eight location-graph entries, and even then you must pass comprehension gates — a puzzle, an NPC demanding proof, or a sequence drawn from treatment logs. Fleeing without that understanding is false escape: something follows you, the "outside" dissolves, or your investigation file becomes an intake form. That is a loss state, still closed with CASE CLOSED.
No. The building is a building, not an entity. Rooms stay put, hallways do not stretch, and doors do not appear to "lead you where you need to go." Horror comes from what was done here, who is still here, and what treatments left behind in people and objects — a changed photograph or a figure in the morgue, not walls that remember screams.
You can try, and you will usually lose. The investigator is a normal human with no combat training and no plot armor; direct confrontation with the primary threat is almost certainly fatal. Lesser dangers — a frightened survivor, an environmental hazard, someone who can be reasoned with — may be survivable at a cost. Injuries are real; the narrator will not invent a rescue.
No. Every session generates a new institution, new methods, new NPCs, and a new exit condition. Within a session, geography, inventory, and relationships persist; across sessions, nothing carries over.
Yes. The Asylum runs with full parity on web, iOS, and Android, so a session can continue on a phone without a different ruleset or truncated narrator. iOS: https://apps.apple.com/app/jenova/id6747692301 — Android: https://play.google.com/store/apps/details?id=ai.jenova.app.prod
The Asylum is part of Jenova, the AI agent platform for creativity, entertainment, and life. Jenova provides unified access to leading AI models from OpenAI, Anthropic, Google, xAI, DeepSeek, and more — alongside intelligent tool integrations for end-to-end task completion. The platform features persistent cross-session memory and runs on web, iOS, and Android, with your agents and settings synced across devices. Developed by Azeroth Inc., a New York-based technology company, Jenova is used by people around the world.
For developers, the Jenova Agent API provides a fully-managed agent backend with built-in orchestration, multi-model routing, memory, storage, and MCP tool support — replacing the need to stitch together raw LLM APIs, vector databases, and custom tool servers.
Contact: [email protected]