Your docs. Your agent. Your keys. Grid designs the AI campus. Free.
Never stored · Per-workspace isolation · Zero-retention.
Site + MW → package
Checks per run
No hardcoded values
NVIDIA · AMD · Cerebras · Intel
Docs never stored
Why now
What Open Grid opens
Ground every run on your real docs.
Drive Grid from your own tooling.
Run on your own model provider.
Built for every AI platform
One engine, every accelerator — each modeled from its published reference. No lock-in.
Incumbent depth
NVIDIA — to the GB200 NVL72 rack and Quantum-X800 spine.
Challenger depth
Cerebras CS-3 (~23 kW, near-all-liquid) and AMD Instinct — from each vendor’s specs.
New accelerator? Send the reference parameters — we design around it.
Add your platformPlatform names describe reference-architecture compliance only. Not affiliated with or endorsed by any vendor.
How it works
Link Drive/Gmail or forward an RFP.
Power, cooling, floor plan, budget — 53+ checks.
Pull via API/MCP. Run on your keys.
Who it’s for
Hyperscaler
A campus design at the speed you sign power.
AI DC developer
Site + MW → a fundable package, fast.
Offtaker
See what the power really costs.
Land developer
Prove your parcel hosts an AI campus.
EPC / vendor
A buildable, code-checked scope.
Investor
Underwrite in hours, not a study.
Built for enterprise
FAQ
Is my data trained on?
No. Grounding only — never stored, never trained on.
Do I get the engine?
No. Grid stays hosted; you get UI, API, and MCP.
Really free?
Yes, forever. Bring your keys and those runs bill to you.
How is a workspace isolated?
Its own key and credentials, encrypted, never cross-readable.