NeuroCognica
System Requirements
Last verified: 2026-07-31
Read this before buying. If your machine does not meet it, the Refund Policy covers you — but it is better for both of us if you know first.
The short version
Windows 10/11 64-bit · 16 GB RAM · an NVIDIA GPU · 60 GB free disk.
You do not need Rust, Python, a compiler, or a GitHub account. The application ships compiled programs. When you consent to the optional setup tasks, the installer is designed to set up the tools it needs; the fresh buyer-install witness remains an active acceptance check, so do not treat this as a completed clean-machine guarantee yet.
Hardware
| Resource | Minimum | Recommended |
|---|---|---|
| CPU | Any modern x86-64 | Intel i7 / AMD Ryzen 7 or better |
| RAM | 16 GB | 32 GB |
| GPU | Works without one, slowly | NVIDIA with 8 GB+ VRAM |
| Free disk | 60 GB | 80 GB+ if you keep a lot of renders |
About the GPU. Blender will render on your CPU and the app will run. But the image-generation paths — the museum wall's painting, the canvas on the easel — load multi-gigabyte models, and on a CPU that is the difference between a couple of minutes and a very long wait. If you do not have an NVIDIA card, buy with that clearly in mind.
About the disk. This is the number people underestimate: roughly 15–30 GB of third-party toolchain, plus about 29 GB of AI image models if you want the painting paths, plus your own renders.
Operating system
Windows 10 (build 18362 or later) or Windows 11, 64-bit. There is no macOS build. Parts of the system run on Linux, but that is a developer path and not what you are buying.
What the installer sets up for you
Only if you leave the dependency step checked, and after showing you the list:
| Tool | Why it is there | Licence |
|---|---|---|
| Blender 4.5 LTS | Builds and renders every scene | GNU GPL |
Ollama + qwen2.5:7b-instruct |
Turns your prompt into a validated scene plan | MIT / Apache-2.0 |
| ComfyUI | Local image generation | GNU GPL v3 |
| Python 3.11, Rust, Git, LLVM, MSVC Build Tools | Runtime and build support those tools need for their own components | Various open source |
Each is installed from its vendor's own official channel, not repackaged by us,
and each stays on your machine if you uninstall Chronos. Full attribution:
THIRD_PARTY_NOTICES.md.
AI image models are separate from the installer binary. The optional installer task is selected by default and, only after the dependency disclosure and your consent, downloads or reuses about 29 GB on NVIDIA or about 9 GB on CPU. Untick it if you will supply the files later. See MODEL_LICENSES.md and WHAT_YOU_GET.md.
Ports
Everything binds to 127.0.0.1. No inbound internet access is required and
nothing listens on your network.
| Service | Port | Required |
|---|---|---|
| Ollama | 11434 | Yes |
| Chronos Director | 7777 | Yes |
| BlenderMCP | 9876 | Yes — the app starts and supervises it |
| ComfyUI | 8000 | Only for the image-generation paths |
The app starts these itself, health-checks them, and restarts them if they die. You should not need to touch them.
Where your work is stored
Under %LOCALAPPDATA%\NeuroCognica\ChronoSophia — renders, .blend files,
the provenance ledger, logs and settings. Uninstalling the application does
not remove them.
How long things take
| With an NVIDIA GPU | CPU only | |
|---|---|---|
| First install | ~1 hour, mostly downloading | same |
| First render of a session | a few minutes — models load once | potentially close to an hour for painted artwork; Create allows 90 minutes |
| Later renders | faster, models stay in memory | still slow |
Before you buy, also read
- WHAT_YOU_GET.md — what the product does, and what it does not
- VERIFY_DOWNLOAD.md — the installer is not code-signed yet; verify your download and expect a Windows warning
- MODEL_LICENSES.md — the AI models you supply
If something does not work, start at SUPPORT.md.
Painted artwork: what your hardware can do
Museum Wall and Canvas-on-Easel paint their picture with an AI image model. Which model you get is decided automatically by what is in your machine, and the difference is worth knowing before you buy.
| Your machine | Model installed | Download | What painting feels like |
|---|---|---|---|
| NVIDIA GPU | FLUX.1-schnell (Apache-2.0) | ~29 GB | Seconds to a couple of minutes, full detail at 1216x832 |
| No GPU, 16 GB+ RAM | Z-Image-Turbo (Apache-2.0) | ~9 GB | Potentially tens of minutes per picture, and noticeably less detailed at 768x512 |
| No GPU, under 16 GB RAM | none | - | Painted artwork is not practical. Everything else still works |
This is not a degraded product on a CPU — it is a slower one with softer
pictures. Geometry, showroom staging, the sealed provenance chain, the hero
render, and the saved .blend are identical on every tier. Only the painted
artwork changes.
Both models are Apache-2.0, so what you make is yours to sell either way, and both run entirely on your own machine with no cloud involved.
If you skip the image models entirely, the recipes that need them say so plainly instead of handing you a blank picture, and every other recipe works normally.