Running a model locally comes down to a question most hardware listings will not answer: does this model, at this quantisation and this context length, fit entirely in this card’s VRAM?
RigForAI works from either end. Choose a model and it returns the GPUs that hold it; choose a chip and it returns the catalogue models that fit. Compatibility is a calculated memory estimate derived from official model configurations — explicitly a capacity question, not a speed ranking.
Once the requirement is known, the same workload can be acquired four ways: a validated bill of materials for a custom build from currently priced parts, exact prebuilt workstation configurations, rented GPU servers from current provider offers, or a cash-flow comparison between building, buying and renting.