About CanIHost
One question, answered precisely: can my computer run this model?
Why the site exists
Every week, thousands of people download a 4 GB model file that their machine cannot load, or buy a GPU that disappoints for local AI, or give up on local models because a first attempt ran at two tokens per second. All of that is preventable with four numbers and a little math: VRAM, RAM, CPU threads, and the model's size.
CanIHost does that math for you — and shows its work. Instead of a vague "should work", you get a fit level, the quantization that matches your memory, an honest (and clearly labeled) speed estimate, and a documented methodology so you can reproduce or challenge every number.
How it works
Recommendations are computed by llmfit, an open-source (MIT) compatibility engine for local LLMs. CanIHost is a web interface for that engine: your profile goes to its API, the analysis comes back, and this site adds editorial context — model pages, GPU pages, and a library of guides that explain the underlying concepts (quantization, VRAM budgets, bandwidth, context memory) without hand-waving.
The quick scan reads your hardware with a visible, read-only PowerShell command — no installation, no automatic upload; you choose what to paste. The site stores nothing about you (see the privacy policy).
How it is funded
CanIHost is free to use and funded by unobtrusive display advertising (Google AdSense), which may be shown on content pages — never inside the checker tool results, and never on error, loading or empty states. Ads never influence recommendations: the model catalogue is static editorial data, not sponsored placements.
Editorial independence
Model and GPU pages state limitations plainly — a 6 GB laptop GPU cannot host 13B comfortably, and we say so even when a retailer's marketing suggests otherwise. When our estimates disagree with community benchmarks, we say that too: see what is estimated versus measured.
Contact
Corrections, suggestions, and hardware results that contradict our estimates are genuinely welcome — they improve the catalogue for everyone. Reach us at contact@exogit.com or via the contact page.