- Direct model loader: run GGUF models straight from disk via llama.cpp's
llama-server, no LM Studio/Ollama needed. One-click binary download
(Vulkan/CPU/CUDA), auto GPU-layer fallback on OOM, Fit-to-GPU VRAM
suggestion, live load progress, and a retractable log panel.
- Audio attachments (Direct backend / mtmd-capable models).
- Video attachments with timestamped frame sampling.
- Edit & resend own messages; chat branching from any prior message.
- Broader image format support via Pillow transcoding.
- Full details in RELEASE_NOTES.md.
RudySen
a7697e4ab4