midudev/canirun.ai

What it solves

It removes the guesswork from running open-weight AI models locally. Users often don't know if their specific hardware (CPU, RAM, GPU) can handle a particular model or which quantization level will provide the best balance of performance and quality without crashing their system.

How it works

The tool uses browser APIs (WebGL, WebGPU, and navigator.deviceMemory) and a lightweight CPU micro-benchmark to detect hardware entirely on the client side. It then compares these specs against a database of 55+ curated models. A scoring algorithm calculates VRAM requirements across seven quantization levels (from Q2_K to F16) and assigns a letter grade (S–F) based on estimated tokens per second, memory headroom, and model size.

Who it’s for

Developers and AI enthusiasts who want to run LLMs, vision models, or coding assistants locally for privacy and cost reasons but need to know which models are compatible with their specific machine.

Highlights

• Zero-install detection: Identifies GPU, VRAM, and memory bandwidth directly from the browser.
• Use-case recommendations: Suggests the best models specifically for coding, chat, reasoning, and vision based on your hardware.
• Detailed compatibility: Provides a full report across 55+ models with specific quantization levels and estimated inference speeds.
• Integration API: Offers a JSON API to check compatibility or get model recommendations for specific hardware profiles.
• One-click setup: Provides installation commands for popular local runners like Ollama, LM Studio, and llama.cpp.