Can I Run AI

Can I Run AI is a free website that estimates which open-weight LLMs a computer can run locally. Users pick or detect a GPU, Mac, or CPU, and the checker returns a fit verdict, memory estimate, and expected speed range for each model in its catalog. It runs in the browser without an account or install, and the site states it never downloads or runs a model itself.

What It Is

It is a local LLM compatibility checker and VRAM calculator. Hardware is detected through WebGPU or WebGL in the browser, or chosen from a profile list, or entered manually (accelerator memory, system memory, bandwidth). The site says raw GPU adapter details are not uploaded. The catalog lists 55 open-weight text-generation models from 12 publishers, including Llama, Qwen, Gemma, DeepSeek R1 distills, Mistral, Phi, and GPT-OSS.

How the Estimates Work

Memory is the sum of model weights, a KV cache sized from each model's architecture and the selected context, and runtime overhead. It is compared with usable memory for the hardware type: discrete GPU VRAM, Apple Silicon unified memory, or system RAM for CPU-only setups. Results are labelled runs comfortably, tight fit, runs with offload, or won't fit. Speed is a tokens-per-second range derived from memory bandwidth, and the docs state it is a planning estimate rather than a benchmark.

Quantization and Runtimes

The checker covers Q4_K_M, Q5_K_M, Q8_0, BF16, FP8, and NVFP4. FP8 and NVFP4 are disabled with a stated reason on hardware that lacks the required NVIDIA architecture. For models that fit, the detail view lists copyable commands for Ollama, llama.cpp, LM Studio, and MLX on a Mac.

Evidence and Rankings

Each model records its source URL, retrieval date, and confidence, favoring official model cards. Where an exact model-version match exists, the list also shows an open-source rank from the public Arena Text Leaderboard dataset. Shared results freeze the catalog version used when the link was created.