Settings

Engines & updates

The installer downloads the right build from the official GitHub repositories (CUDA for NVIDIA, Vulkan for AMD/Intel, or CPU-only) and unpacks it. Each engine has its own directory and update button:

• llama.cpp — the text server (detects the installed version and newer releases)
• stable-diffusion.cpp — images and video
• whisper.cpp — speech-to-text for voice mode
• KVMem (experimental) — an alternative engine

Hardware detection

Settings show processor, graphics card, VRAM, driver, CUDA version and RAM (via nvidia-smi, with a WMI fallback). VRAM from WMI can be inaccurate for cards above 4 GB — install NVIDIA drivers for an accurate reading.

AI integration

Choose a provider (Claude, OpenAI, Gemini) and enter an API key. It is stored locally, encrypted, and sent only directly to the chosen provider. Auto-tune and the benchmark quality judge use it.

Backup & language

Export saves profiles, per-model flags, favorites and the API key to a single file — handy before reinstalling. Import overwrites current settings. The interface language (Polish / English) defaults to your OS language.