skycontact/skyvision.cpp is a local vision-language inference engine for asking questions about images and videos and getting answers. According to the GitHub repository description and README, it provides a unified interface that selects the GGUF backend matching the model architecture. GGUF is a file format for storing and distributing models. The repository ships a command-line tool, a web interface, and a Model Context Protocol (MCP) server that agents can connect to. No release is registered, so there are no new versions or changes to confirm this time.
The core of this project is handling multiple vision-language models within a single engine. The description lists moondream3.1, VideoChat3, Qwen2.5-VL, MOSS-VL, VideoMAE/ViViT and SmolVLM as supported targets. Some models also support video input, or classification and feature-extraction modes. Video question answering and position estimation, and video classification, are split across separate tools. The repository's primary language is C++, and it introduces itself as not requiring Python or the CUDA toolkit at runtime.
However, the scope of support and compatibility differ by backend, and also depend on model conversion and runtime conditions. Therefore, it cannot be assumed that all models work in the same way based only on the listed model names. The repository's license is listed as NOASSERTION, so usage terms cannot be judged from this information alone. Since it is an early repository with no registered release, compatibility and license terms for the models you need should be checked separately before applying it in practice.