Free · Private · Runs in your browser
Which AI models can your PC run?
We detect your GPU and memory and rank the open models that actually fit — chat, coding, image and video. Nothing is uploaded; everything is computed on your machine.
Works best in Chrome or Edge on desktop. Apple Silicon, NVIDIA, AMD and Intel supported.
All results are estimates. Real performance depends on drivers, cooling, background apps and OS memory pressure. Use the grades as a guideline, not a guarantee.
How it works
Open the page
Visit in a desktop browser. No account, no extension, no install.
Detect or pick your hardware
Your browser reports the GPU, RAM and CPU cores. Prefer manual? Choose your device from the list.
Compare and run
Review the top open models that fit, then grab the GGUF from Hugging Face and run it with Ollama, LM Studio or llama.cpp.
Why run AI locally?
Privacy
Your prompts never leave your device. No cloud account, no logging, no third-party review.
No per-token cost
Once the model is downloaded, inference is free. No subscriptions, no rate limits.
Works offline
Flaky connection or no internet at all — local models keep working anywhere.
Full control
Open weights mean you own the model. Fine-tune it, serve it, or keep it forever.
Cloud assistants are still stronger on the hardest tasks. AI Can Run is for the open-weight models you can download and run yourself.
Common questions
How much VRAM do I need?
As a rule of thumb: 8 GB runs 3–9B chat models at Q4; 12–16 GB covers most 12B–27B models; 24 GB opens 30B dense models and local image or video generation. Mixture-of-experts models load all experts into memory, even though only a few are active per token — so a 128B MoE still needs 80 GB+.
Is my data sent anywhere?
No. Hardware detection runs entirely in your browser using WebGL, WebGPU and standard navigation APIs. The page is fully static — nothing is uploaded to any server.
What is quantization?
Quantization stores model weights in fewer bits so files are smaller and use less memory. Q4_K_M is the usual sweet spot for local chat — much smaller than full precision with only a modest quality drop. Higher formats like Q8 or F16 look closer to the original but need more VRAM.
Which runtime should I use?
Ollama and LM Studio are the easiest starting points. llama.cpp gives you the most control and runs on almost anything, including Raspberry Pi. Any of them can load the same GGUF files.
Can I use this for a business?
That depends on the model licence, not on this tool. Models carry their own terms (Apache 2.0, MIT, Llama Community, etc.) — check the licence on each model card before commercial use.