What does it take to run AIBuild?
The estimator core is a lightweight desktop app — any modern office PC will do. The AI features run open-weight models locally, and that's where the GPU matters. Pick the column that matches the plan you want to use.
Core estimator only
Planning, catalog, estimates and exports. Everything except the AI toolkit.
| Component | Minimum | Recommended |
|---|---|---|
| OS | Windows 10 64-bit | Windows 11 64-bit |
| CPU | Any dual-core, 2 GHz+ | Quad-core (2020 or newer) |
| RAM | 4 GB | 8 GB |
| GPU | Any — integrated graphics is fine | Any |
| Storage | 500 MB free | 2 GB free (projects + backups) |
| Network | Only needed to fetch supplier product pages — the app itself works fully offline | |
If your PC can run a browser with a few tabs, it can run AIBuild Basic.
Estimator + on-device AI toolkit
AI product extraction and photo-to-3D generation, all running on your own hardware. No cloud, no per-use fees.
| Component | Minimum | Recommended |
|---|---|---|
| OS | Windows 10 64-bit | Windows 11 64-bit |
| CPU | 6 cores (Ryzen 5 / Core i5) | 8+ cores (Ryzen 7 / Core i7 or better) |
| RAM | 16 GB | 32 GB (64 GB for heavy multitasking) |
| GPU — AI extraction | NVIDIA, 8 GB VRAM (small model, slower) | NVIDIA RTX 4070 Ti / 5070 class, 12–16 GB VRAM |
| GPU — best experience | — | RTX 4080 / 5080 / 5090 class, 24–32 GB VRAM (runs the full-quality 27B model) |
| GPU — photo → 3D Beta | 12 GB free VRAM while generating (the AI extractor pauses automatically — AIBuild coordinates both on one GPU) | |
| Storage | 25 GB free (NVMe strongly advised) | 40 GB free NVMe — models load in seconds, not minutes |
| Network | One-time model download (~20 GB). After that, fully offline. | |
Only NVIDIA GPUs with CUDA are supported for local AI today. AMD / Intel GPUs and Apple Silicon can run the Basic tier; the managed cloud AI option (coming soon) will bring Advanced to them.
Advanced features, hosted
Advanced plans will be able to offload the AI work to our managed inference service instead of a local GPU. You'll get every AI feature on ordinary hardware — extraction and photo-to-3D will run in the cloud instead of on your desk.
Any hardware
The same modest spec as the Basic tier — the heavy models run on our side.
Your choice per feature
Mix and match: local extraction for privacy, hosted photo-to-3D for speed — switch any time.
Local when you're ready
Add a CUDA GPU later and the same plan flips to fully local, fully private inference.
Ready when your hardware is.
Start on Basic today, turn the AI on whenever it makes sense.