One click still contains a lot of engineering
Paste a Hugging Face repository, let Capix inspect its immutable metadata, and receive a quote for compatible live capacity. After confirmation, the deployment saga reserves funds, allocates a GPU, starts the appropriate serving runtime, downloads the model and waits for an OpenAI-compatible health check.
Capix supports common checkpoint layouts including safetensors, GGUF and PyTorch binary shards. The format and model metadata influence which runtime and minimum VRAM class are eligible.
Private means account-scoped
A healthy private endpoint is discoverable only by its owner. It appears in the owner’s model catalog and can be used from web chat, Capix IDE, Capix Code or the OpenAI-compatible API. Destroying the deployment removes it from the active catalog and stops future usage charges.
Provisioning normally takes several minutes because model weights must reach the selected host and the endpoint must answer a real health check before Capix calls it ready.