Local inference / Ryzen AI Max

Local LLM inference, hand-tuned for one chip. The Ryzen AI Max+ 395.

Install

GPU

Run

Demo

2.4× faster than llama.cpp on nginx log triage through Pi. Same Ryzen AI Max+ 395, same Qwen3.6-35B-A3B UD-Q4_K_XL, one engine at a time.

Web UI

No commands to memorize.

snowllm webui

Download models, chat and monitor inference in your browser. Web UI guide ↗

Integrations

snowllm pi --port 8000

Automatically generates ~/.pi/agent/models.json for pi.dev.

Other tools connect through APIs compatible with OpenAI and Anthropic. Integration guide ↗

Run

snowllm pull qwen3.8-flash-next-q3-k-xl
snowllm qwen3.8-flash-next-q3-k-xl