llama.cpp creator Georgi Gerganov joins Hugging Face to keep local AI's engine running

The world’s most critical piece of local AI plumbing just got a much more stable future. Georgi Gerganov, the creator of the ggml library and the lead maintainer of llama.cpp, is joining Hugging Face along with his team. This isn’t an acquisition of a project — llama.cpp remains 100% open-source and Gerganov retains full technical autonomy. Think of it more like Hugging Face becoming the project’s long-term benefactor and institutional support system.

The announcement frames the move as a natural next step in a years-long collaboration. Hugging Face already employs key llama.cpp contributors like Son and Alek, so the teams are deeply intertwined. Gerganov and his crew will continue to dedicate all their time to the project, but now with sustainable resources backing them. The goal is to remove the existential risk that often haunts critical solo-maintained infrastructure, ensuring the software doesn’t burn out its creator.

On the technical side, the vision is a seamless pipeline between the two foundational layers of modern AI. Hugging Face’s transformers library is the “source of truth” for defining model architectures, while llama.cpp is the engine that runs them efficiently on consumer hardware. The plan is to make shipping a new model from a transformers definition into a quantized, locally-runnable ggml format a near “single-click” affair, stripping away the current friction.

The larger ambition is blunt and worth taking seriously: building the ultimate inference stack to make open-source superintelligence accessible on everyday devices. As local inference matures into a genuine competitor to cloud APIs, the focus will shift to packaging and user experience, making it trivial for a casual user to deploy a powerful model. The bet here is that the future of AI won’t live in a data center you rent by the hour, but on the laptop in front of you — and this deal ensures the software layer for that future has a fighting chance.