ggml.ai, the team behind llama.cpp, joins Hugging Face

Background

ggml.ai was founded in 2023 with the mission of supporting the development and adoption of the ggml machine learning library ^[2]. That library is closely tied to llama.cpp, the project built by the same founding team; both are widely used for running models locally on consumer hardware ^[1]. Running a model locally means the computation happens on hardware the user owns, such as a laptop or a desktop machine, instead of being handled by a remote service. For the people who rely on these tools, the practical questions are how much hardware a model needs and how easily it can be set up, and that is the ground both projects compete on.

It helps to separate the two layers. A machine learning library supplies the low-level pieces — numerical routines, memory handling, the data structures that hold model weights — that other programs build on. A program such as llama.cpp uses those pieces to load a model and run it. Improvements at the lower layer therefore reach everything built above it, which is one reason the direction of the library matters beyond any single application.

The announcement

On 20 February 2026, Georgi Gerganov announced that ggml.ai, the founding team of llama.cpp, was joining Hugging Face ^[1]. The framing was long-term: Hugging Face would provide long-term resources for the work, while the projects stay 100% open source ^[1].

What stays the same

The announcement drew a line between backing and control. The community would continue to operate autonomously and make its own technical decisions, and the team would keep working full-time on ggml and llama.cpp ^[1]. The people maintaining the code, the process for deciding what goes in, and the open-source terms are all meant to stay as they are; what changes is who carries the cost of the work continuing.

What changes

The stated goals were about making the software easier to reach: "single-click" integration with the Transformers library, and easier local deployment ^[1]. The team said extra focus would go on improving user experience and on integration with the Transformers library ^[2]. The two organizations were not strangers — the announcement noted that Hugging Face engineers had already contributed several core functionalities to ggml ^[2] — so the integration builds on work that predates the move.

Why it matters

A small team maintains low-level infrastructure that a large number of other projects depend on. A library such as ggml sits underneath the applications people actually use, so its health is a shared concern even for users who never touch it directly. Bringing such a team inside a larger organization can supply stability, but it also raises the question of whether the project will keep serving the users who chose it for being independent and open. The terms described here — community autonomy over technical decisions, full-time work on the same projects, 100% open source, and long-term resources — are an attempt to answer that question in advance. Whether the arrangement holds will show up in the code and the pace of releases over the following years.