I shipped a windows desktop app for running local LLMs with a button that turns your "no thats wrong" into actual LoRA training data

In three linesSEELS, a Windows desktop app for local LLMs, lets users correct model replies via a « Teach » button that accumulates corrections into a JSONL corpus, then triggers PEFT LoRA fine-tuning without terminal access. Includes local STT/TTS (Whisper/Piper), hardware dashboard, 0.6B model pre-trained on 110 examples. Free stable version; pro tier (image/video gen, MCP) and max tier (workflows, multi-GPU) in roadmap.

Summary generated by Claude — human-verified