About CanMyPCRunAI
Working out whether your computer can run a particular local AI model usually means reading forum threads, guessing at quantization names, and downloading several gigabytes to find out. This site answers the question before you download anything.
Maintained by Delcio Pedro
I maintain CanMyPCRunAI and its hardware detector. The goal is to make model memory requirements understandable before a large download. The checker estimates compatibility; its results are not independent hardware benchmarks.
The detector source is public under the MIT license. You can inspect how it collects hardware and report a reproducible problem. Read a worked calculation and its limitations.
How it works
The Windows executable or macOS/Linux Python detector reads CPU, RAM, GPU, VRAM, available acceleration backends and free disk space, then sends that profile to the scan session you started. A deterministic engine compares it against every model in the catalog and explains which ones fit, which need part of the model offloaded to system RAM, and which do not fit at all.
Windows users run the executable. Mac and Linux users need Python 3.9 or newer and Terminal; the scan page supplies the paired command. You can also use a labelled reference estimate or enter hardware manually. The Windows executable is currently unsigned; the free SignPath application is pending, and Windows may display a publisher warning.
Where the numbers come from
Model sizes, quantizations, context windows and licences are read from the official Ollama registry and model pages — not estimated. Memory requirements are calculated from model weights, KV cache at your chosen context, compute buffers and runtime overhead, minus a reserve for your display and operating system.
Where a figure is not published, the site leaves it blank and lowers the confidence it reports, rather than presenting a guess as fact. There are no benchmark numbers on this site, because we have not measured any.
What it deliberately does not do
- No accounts, no sign-up, no email address
- No IP addresses stored
- No reading of your files, history or credentials
- No invented figures when data is missing
The full detail is in the privacy policy.
Corrections and contact
Report the model tag, chosen context, operating system, expected result and a source supporting the correction through the public GitHub issue tracker.
Do not include a pairing code, scan link with tokens, credentials or private files. An issue is a public report; review hardware details before sharing them. Corrections are checked against the model or hardware publisher and the calculation is rerun before a verification date is updated.