README.md
Rama main · réplica de lectura del repositorio original.
<div align="center">
# 🐆 Panther Minor
### Self-hosted AI workstation stack for [AMD](https://www.amd.com/en.html)
A reproducible, self-hosted AI cluster tuned for **AMD Ryzen + RDNA 4** systems: **llama.cpp**,
**stable-diffusion.cpp**, **Open WebUI**, **Prometheus** and **Grafana**, with **Tailscale access**, **hardened SSH**
and **smart GPU usage management**.
**[📚 Documentation](docs/README.md)** · [Installation](docs/installation.md) · [Operations](docs/operations.md) ·
[CLI](docs/cli.md)
</div>
---
## ✨ Highlights
| Feature | What it gives you |
| --------------------------- | ------------------------------------------------------------------------------------------------------------- |
| **Local inference** | OpenAI-compatible LLM API powered by [llama.cpp](https://github.com/ggml-org/llama.cpp) |
| **Local image generation** | OpenAI-compatible image API powered by [stable-diffusion.cpp](https://github.com/leejet/stable-diffusion.cpp) |
| **Built-in monitoring** | Prometheus, Grafana, and exporters for host + GPU visibility |
| **GPU power & VRAM saving** | Idle model unloading and serialized large-model switching |
| **Secure remote access** | Tailscale-scoped ports, key-only SSH on port `2222`, firewall, and fail2ban |
| **Reproducible setup** | One CLI-driven installation flow for the whole workstation |
> [!IMPORTANT]
> Panther Minor is a **single-tenant** system for one trusted user and workload. It is not suitable for multi-user or
> untrusted environments.
## 🚀 Quick start
**Requires** Ubuntu Server 26.04 LTS+, 2x AMD RDNA 4 GPUs, a Tailscale account and a domain — see
[Installation](docs/installation.md#-prerequisites) for hardware and BIOS settings.
```bash
# 1. Install the latest release into ~/panther-minor and prepare the host (reboot afterwards)
curl -fsSL https://github.com/rozsival/panther-minor/releases/latest/download/install.sh | bash
cd ~/panther-minor
sudo ./bin/panther-minor setup
sudo reboot
# 2. Reconnect on port 2222, join Tailscale, scope ports to it
ssh -p 2222 <user>@<server-ip>
sudo tailscale up
sudo ./bin/panther-minor setup env
# 3. Issue TLS certificates and enable renewal
./bin/panther-minor proxy certbot --domain <domain> --challenge-record _acme-challenge
./bin/panther-minor proxy setup-cron
# 4. Download the default models and start the cluster
./bin/panther-minor models llm download Qwen3.8-27B
./bin/panther-minor models llm download Qwen3.5-2B
./bin/panther-minor models llm download Qwen3-Embedding-0.6B
./bin/panther-minor cluster start
```
Then open `https://<domain>:8080` (Open WebUI) or point an OpenAI client at `https://<domain>:8000/v1`.
> [!WARNING]
> After `setup`, SSH accepts **keys only on port 2222**. Keep your current session open until a second one connects.
## 📚 Documentation
Everything else — architecture, networking, models, GPU management, monitoring, coding harnesses, CLI and
development — lives in **[docs/](docs/README.md)**.
## 👤 Ownership
| Item | Details |
| ---------- | ------------------------------------------------------------------------------------------------------- |
| Maintainer | [@rozsival](https://github.com/rozsival) (see [`CODEOWNERS`](CODEOWNERS)) |
| Issues | [GitHub Issues](https://github.com/rozsival/panther-minor/issues) |
| Companion | [Panther Minor Controller](https://github.com/rozsival/panther-minor-controller) — remote power control |
| License | [MIT](LICENSE) |