Documentation
Short and practical. CozyGPU runs on your machine, inference in your own RunPod account; most questions are about that boundary.
- SetupInstall, add keys, run the first job. What the app creates in your RunPod account.Read
- KeysWhere to get the RunPod API key, the RunPod S3 API key and the optional xAI key, and where they live.Read
- Custom LoRAsDrag and drop .safetensors files, trigger words, strength, storage and sync.Read
- LicenseActivation, one device per key, moving it, the offline grace period, renewal and cancellation.Read
- TroubleshootingGatekeeper, SmartScreen, the first-run pull, RunPod balance and spend limits, idle endpoints.Read
How it fits together
CozyGPU is a desktop app. On your computer it keeps a library (images, videos, prompts), your settings and your keys. When you generate, it talks directly to RunPod with your key: it creates a serverless endpoint per model in your account (named cozygpu-<model>), submits the job, polls it, and downloads the result into the library.
There is no CozyGPU server in that path. The app contacts us only to check the license and the update feed. The security notes list every outbound request.