
Colibri
Frontier MoE models on hardware you already own — pure C, zero dependencies.
About
Colibri is a tiny inference engine (pure C, zero dependencies) that runs frontier Mixture-of-Experts models on ordinary hardware by streaming experts from disk — a 744B-class model becomes usable on a laptop with ~10GB RAM. The reel called it "Calibri"; the real name is Colibri.
git clone https://github.com/JustVugg/colibri
# follow the repo's build instructions (pure C — quick build)More in Local & Free AI
View allOllama
Run open models like Llama, Qwen and DeepSeek locally with one command.
ollama pull llama3.1
Local & Free AIOpen sourceActive186.4K
AnythingLLM
Private, local-first app to chat with your documents and run AI agents
Local & Free AIOpen sourceActive68.1K
Jan
A ChatGPT-style app that runs 100% offline on your computer.
Local & Free AIOpen sourceActive45.6K