# Paddock documentation

Install the current alpha, understand the trust boundary, and bring one local model online.

## Start with the support boundary

The current public channel is an honest Linux x86-64 alpha. Review the release evidence before changing a host.

- llama.cpp is the supported production runtime.
- Current real-runtime evidence covers one Linux amd64 CUDA RTX 5090 host.
- CPU, ROCm, Vulkan, arm64, and experimental vLLM paths are not implied by the alpha label.

## Install

The check command verifies the current download without modifying the host. Run it before installation.

```text
bash <(curl -fsSL https://get.kinra.ai/paddock/install.sh) --check
bash <(curl -fsSL https://get.kinra.ai/paddock/install.sh)
```

## Build the managed runtime

Paddock detects the host, checks out its known-good llama.cpp revision, and builds the appropriate backend.

```text
sudo -u paddock PADDOCK_ROOT=/var/lib/paddock \
  /usr/local/bin/paddock install-llamacpp
```

## Pull a model

A pull stages, validates, and atomically registers the selected GGUF artifact set.

```text
sudo -u paddock PADDOCK_ROOT=/var/lib/paddock \
  /usr/local/bin/paddock pull unsloth/Qwen3-30B-GGUF:Q4_K_M
```

## Operate locally

Start the service, then use the loopback dashboard and API.

```text
sudo systemctl enable --now paddock.service
# dashboard and API
http://localhost:5467
```

> Do not expose the listener directly to an untrusted network. Use a private network or an authenticated reverse proxy when remote operation is required.

## Maintained depth

- [Release and compatibility](https://get.kinra.ai/paddock/)
- [Installer](https://get.kinra.ai/paddock/install.sh)

## Source authority

The Paddock repository and the release publication at get.kinra.ai/paddock. Product behavior, release state, and full operating detail remain authoritative there.