Local inference.
One chip, hand-tuned.

Local LLM inference on Ryzen AI Max APUs, optimized for low concurrency.

01Install SnowLLMUse the installer on Linux or WSL2; install manually on Windows.↗02Run your first modelDownload a checkpoint, open the Web UI and start chatting.↗03Connect your toolsUse Pi or the OpenAI and Anthropic compatible APIs.↗

Requirements

Linux, Windows 11 or WSL2 with a gfx1151 APU. Tested on Ryzen AI Max+ 395; other Ryzen AI Max 300/400 APUs remain unverified.

Installation requirements · Supported models

Guides

• Web UI: download, load and chat.
• Performance tuning: memory, context and speculative decoding.
• CLI options: flags and defaults.
• Configuration: storage and environment variables.
• Usage statistics: collection fields, reporting behavior, and opt-out.

Benchmarks · Discussions