Documentation

Everything you need to go from a one-line idea to a published, playable game.

Quickstart

  1. Create a dev account. Sign up with email or Google. Players are free; building uses credits.
  2. Describe your game. In the studio, type a prompt like a cozy island fishing sim with day/night. Be as specific as you like about look, loop, and mood.
  3. Approve the plan. BLACKRAT turns your prompt into a plan you review line by line — 👍 the rows you like, 👎 the ones you don't, and the agent revises until it's right.
  4. Play the demo. A full, playable AAA demo build proves the real look and feel — a well-iterated hero, real textured ground, a polished controller. Play it, then thumbs-rate everything in the scene.
  5. Generate the full game. Happy with the demo? Generate the complete game — real meshes, materials, animation, audio, world FX, and UI.
  6. Publish. Push it to the Play library or share privately by link.

How it works

BLACKRAT is an AI game engine built around an approval-driven, multi-agent pipeline. The flow is always the same five stages, and you stay in control at every gate:

📝

Prompt

Plain-English description of the game you want — genre, look, mechanics, mood.

Approve plan

The prompt becomes a structured plan. You approve or reject each row; the agent iterates on rejections.

🤖

Multi-agent build

Specialist agents generate 3D meshes, PBR materials, rigs + animation, audio, world FX, netcode, and UI in parallel — each gated to an AAA quality bar.

🎮

Play-test

You play a demo build in the browser and rate it. Feedback feeds back into a revise loop.

🚀

Publish

Ship to the public Play library, share by private link, or export to self-host.

Under the hood, the engine treats each core (Noun, Verb, Adjective, Preposition, Conjunction, and the World/FX suite) as a rubric that directs the build and an un-gameable auditor that gates it. Generation runs in isolated, single-use sandboxes; your project state is captured as immutable, content-addressed versions so history can be reviewed, reverted, and cherry-picked — never silently rewritten.

Pricing & credits

Playing games is free. Building consumes credits, because every generation makes real model and GPU calls. Credits are priced on the underlying model + GPU cost plus a clear margin — no hidden multipliers.

Publishing & ownership

You own the content you create. Publishing grants BLACKRAT a license to host and distribute what you choose to share — nothing more.

FAQ

Do I need to know how to code or use a game engine?

No. You describe the game in plain English and approve a plan. There is no engine to learn — the multi-agent build produces the meshes, materials, animation, audio, world FX, and UI for you. You can still push back on any plan row to steer the result.

Is it free to play games built with BLACKRAT?

Yes — playing is completely free in your browser. Only building consumes credits, because generation makes real model and GPU calls. There is no free dev tier, but players never pay.

Who owns the games and assets I create?

You do. You retain ownership of everything you create. When you publish, you grant BLACKRAT a license to host and distribute it; you can unpublish, export, or request deletion at any time.

Can I export my game and host it myself or sell it elsewhere?

Yes. You can export your project bundle to self-host or take it to other stores, and you can sell your individual assets on the BLACKRAT marketplace. Your project data is exportable and deletable on request.

How do credits work and can I avoid overspending?

Credits are priced on the real model + GPU cost with a clear margin. Each build job places a pre-flight hold and is recorded in an idempotent ledger, so you can never be charged twice or driven negative. Subscription tiers include a monthly allotment; overage and auto-recharge cover the rest.

What if the AI gets the game wrong?

You approve the plan before any build runs, and you play a full, playable AAA demo before generating the full game. Thumbs-down feedback on plan rows or scene elements feeds a revise loop, so you converge on what you actually want instead of accepting a one-shot result.

Ready? Create an account or see what's new →