companies

ggml.ai

The small Sofia-based lab behind llama.cpp and whisper.cpp, the C/C++ engines that let LLMs and speech models run on ordinary laptops and phones without a GPU.

Georgi Gerganov founded ggml.ai in Sofia, Bulgaria in 2023 to keep building on ggml, the tensor library he wrote to run OpenAI's Whisper on a MacBook without Python or CUDA installed. That side project became llama.cpp, now one of the most widely used pieces of open-source AI infrastructure: a C/C++ inference engine that quantizes model weights into the GGUF format and runs them on CPUs, Apple Silicon, and consumer GPUs alike. The company was funded pre-seed by Nat Friedman and Daniel Gross specifically so Gerganov could work on ggml full-time. Beyond llama.cpp, the team maintains whisper.cpp, a from-scratch C/C++ port of Whisper that runs speech-to-text on a Raspberry Pi or an iPhone, and the underlying ggml library both projects share. In February 2026, ggml.ai was acquired by Hugging Face, which said Gerganov's team would keep full technical leadership over the projects and continue developing them under the same MIT license. llama.cpp crossed 100,000 GitHub stars in March 2026, a pace commentators compared favorably to how long it took PyTorch and TensorFlow to reach the same mark, underlining how central the project has become to running open-weight models locally.

Founded

2023

Headquarters

Sofia, Bulgaria

Sector

infrastructure