Flux by Finn

Flux v0.1.9 introduces:

• Gemma 4 models — Flux Steady (Gemma 4 E2B) and Flux Smart (Gemma 4 E4B) replace previous models with native vision support and mmproj auto-download
• Vision for all models — Flux Lite (Qwen 3.5 0.8B), Steady, and Smart all support image attachments via paperclip button
• Image memory — previously uploaded images are remembered in conversation history
• Web Search currently doesn't work - will be fixed on the next update
• Proactive context compaction — automatically summarizes older conversation turns before the context window fills up
• Maximum speed — doubled batch sizes (4096), GPU offload for all models, per-model context tuning, 60fps streaming
• Thinking process UI — Gemma 4 reasoning rendered as a collapsible section with opacity transitions
• Real desktop RAM and storage detection via native system calls (sysctl/df/PowerShell)
• Model warmup on app start for instant first message
• Lower RAM requirements — Lite 3GB, Steady 5GB, Smart 7GB

We hope you enjoy this release!