v2.0.0-rc.12

HomeGenie Server 🧞 v2.0.0-rc.12

This release candidate introduces a massive overhaul to the Lailama Local AI engine and debuts a new, robust Async Download Manager.

🤖 Lailama (Local AI) Overhaul

The local intelligence module has been rewritten to be more stable, efficient, and user-friendly.

• Next-Gen Model Support: The model registry (models.yaml) now defaults to late-2025 SOTA models, including:
  • Qwen 3 & Gemma 3
  • Llama 3.2
  • GPT-5.2 & DeepSeek 3.2 Distills
• Download Management: Fully manual workflow implemented. You can now Start, Pause, and Resume massive GGUF model downloads directly from the settings UI.
• Smart Stream Buffering: The engine now surgically removes generation artifacts (e.g., <|end|>, /avatar) and hallucinations (e.g., User:), ensuring a clean chat experience.
• Performance & Stability:
  • Fixed memory leaks via robust resource disposal and safe model warmup.
  • Hardened System Prompt to suppress internal monologue.

⚡ Widget & UI Improvements

• Optimized Rendering: Implemented 50ms throttling for the chat widget, drastically reducing CPU usage during generation.
• UX Enhancements: Added Smart Auto-Scroll (won't disturb you if you are reading history), a CSS-based terminal cursor, and a new "Landing UI" for pending downloads.

🛠️ Core Infrastructure

• Async Download Manager: A new core service handling HTTP(S) file transfers with full support for Range headers (pause/resume) and temporary .part file handling.

🐛 Bug Fixes & Misc

• Fixed a resource leak in Widget Editor when updating component views.
• Improved navigability in the HomeFlix (Media Server) UI.

Full Changelog: v2.0.0-rc.11...v2.0.0-rc.12