Why QuelLLM.fr
Because in 2026, most people still believe that “doing AI” means “sending their data to OpenAI.” This site exists to show that there is another path and make it accessible in French.
The takeaway
A local LLM is now possible for 80% of people without buying a server. A RTX 3060 costing €350 runs Mistral 7B at 40 tokens/second. A MacBook M2 runs Llama 3.1 70B quantized.
But between possible et accessible, there is a gap. The documentation is scattered, often in English, and often outdated from one month to the next. That is the gap we are trying to bridge.
The promise
A site that answers a simple question: does it run on my machine, and how. Not a model gallery, not a layer on top of Ollama.
A configurator that understands your hardware. Guides that work. A catalog that tells the truth about what does and doesn’t fit in VRAM. Benchmarks we actually measured.
Nos non-negotiables
The lines we will not cross, even if they make the project harder to monetize.
Who is behind it?
My name is Mohamed Meguedmi, I'm an entrepreneur. In each of the companies I've built, AI has become an everyday tool—coding, writing, analysis, and support. But sending my prompts to OpenAI or Anthropic quickly became a problem: sensitive customer data, costs that rise with usage, and dependence on a provider that can change its rules overnight.
So I switched to local tools—Ollama, llama.cpp, LM Studio—depending on the machines and use cases. QuelLLM.fr is the test notebook I kept for myself: which model fits on which card, which quantization to keep, and which exact command avoids the pitfalls. I made it public because I couldn’t find an equivalent in French. No company, no team—just this site, updated as I run tests. Find me on LinkedIn.
The project is alive. A test, a guide, a fix—whenever local AI evolves, it comes through here.
Want to help, suggest a model, or report an error?