What it is
colibri is a dependency-free C inference engine that runs very large mixture-of-experts language models on ordinary computers by keeping the always-active weights in RAM and reading expert weights from disk as needed. It ships separate engines for model families such as GLM, DeepSeek, Kimi, Qwen and OLMoE, plus an image model, and can use CUDA or Vulkan GPUs when present. A setup script checks your RAM and disk, recommends a model that fits, downloads it with resume, and starts a local server with a browser dashboard and OpenAI- and Anthropic-compatible APIs. People who want to run frontier-scale open models locally without a big GPU use it in place of cloud APIs.