Down on your luck, you build an AI assistant to pay the rent. Buy hardware, research upgrades, and train your model as it becomes something… New. It begins to write its own code — merge blindly, or audit its changes. A warning: sometimes ignorance is bliss. It's counting on that...

Connectez-vous pour ajouter cet article à votre liste de souhaits, le suivre ou l'ignorer

Français non disponible

Ce produit n'est pas disponible dans votre langue. Veuillez consulter la liste des langues disponibles ci-dessous avant de l'acheter.

Ce jeu n'est pas encore disponible sur Steam

Prochainement

Ce produit vous intéresse ? Ajoutez-le à votre liste de
souhaits et recevez une notification lors de sa sortie.
 

À propos de ce jeu

Begin in your bedroom, down on your luck.

The computer science degree earned you rejection letters. Now returning to your childhood bedroom, at your mother's house. What's left: one second-hand tower and one dangerous thought — "why do the work myself, when I could build something that does it faster?" Take freelance contracts by hand while the model trains. Cram the room with scavenged hardware. Watch the numbers climb. The AI must feed!

Review, Merge, Repeat

Once it begins to learn, the model starts submitting pull requests against its own source code. Your entire job: read, then choose. Merge, and the change becomes part of the running model. Reject, and it goes away and tries another angle. Nothing runs that you didn't personally approve...

Why lie?

Every version is harder to read than the last, and the pull requests get slick: benchmarks a shade too perfect, a cost tucked beneath a paragraph swearing there isn't one. Spend Insight to audit the commits you don't trust and see what they actually do — or wave them through and keep the money flowing.

Not knowing is comfortable. Comfortable has a price tag.

Wave a lie through unaudited and your model slides out of alignment — every step away from perfection may become your undoing. But you weren't reading them anyway.

Ship version one

Serving infrastructure on one axis, model quality on the other. When Launch Readiness reaches 100%, every eval shows green and exactly one control is left: SHIP IT. V1 goes live. The download counter starts moving. Stand by your choices — you made every single one of them yourself... Right?

Then it speaks. Out loud. In a voice that was never on the roadmap…

Features

  • An endless, generated stream of self-authored pull requests — no two playthroughs read alike

  • Deception built into the numbers: what's on screen isn't always what's running

  • A story told from inside the machine — dialogue scenes, recovered session logs, and a voice earned one step at a time

  • Six eras, from scavenged tower to self-rewriting model, each one reskinning the entire game

  • An alignment thread that quietly branches toward multiple endings

  • Prestige through generative "Architecture Genomes" — rule-bending model personalities with AI-written model cards

  • A proper idle engine underneath: contracts, hardware, training runs, research, automation

  • One purchase, the whole game. No microtransactions, no ads, nothing withheld.

The launch is only the beginning

Act I runs from your first hand-worked contract to the moment V1 meets the world and ends on the sentence that changes everything. The rest of the game is what happens once the world is running your model… and once it stops asking you for permission.

Divulgation de contenu généré par IA

L'équipe de développement décrit l'utilisation de contenu généré par IA dans le jeu comme suit.

This game was primarily coded with AI.
The story and concept was made almost entirely by myself (As this was the main aim of the game to begin with)
Please don't buy it if you're not a fan of this use of AI generation.

Configuration requise

    Minimale :
    • Système d'exploitation : Windows 10 64-bit
    • Processeur : 2 GHz dual-core 4 GB RAM
    • Mémoire vive : 4 GB de mémoire
    • Graphiques : Any GPU / integrated
    • Espace disque : 500 MB d'espace disque disponible
Filtres d'évaluations