RigCheck Local Model Fit Checker

Check if your GPU or Mac can run a new model using real file sizes and KV cache

Problem

Every time a new open-source model drops, the community estimates VRAM by parameter count times bit width, which often diverges significantly from real requirements. Information is scattered and lacks credible sources.

Solution

Provide a checker per model that uses actual repository file sizes and computes KV cache layer by layer following llama.cpp allocation (including sliding window, linear attention, shared KV layers), then compares against available device VRAM/RAM. Each number is labeled by source type (official, community, estimate, measured).

Users

Developers deploying open-source LLMs locally, AI enthusiasts, small research teams

MVP

Support mainstream open-source models and common GPU/Mac memory configs; input a model and device to output feasibility and required VRAM/RAM; label every number with its source; provide pages browsable by GPU, Mac memory, and VRAM tier

Revenue

Free basic lookup; paid subscription for advanced filters, batch comparisons, and measured data reports; affiliate promotion with cloud GPU rental providers

Why now

Open-source models iterate extremely fast, local running demand is strong, but accurate assessment tools are scarce. An indie developer can build a moat with measured data.